Linear Lie Groups, Volume 35

LINEAR LIE GROUPS

This is Volume 35 in PURE AND APPLIED MATHEMATICS A Series of Monographs and Textbooks Edited by PAULSMITHAND SAMUEL EILENBERG A complete list of titles in this series appears at the end of this volume

LINEAR LIE GROUPS

HANS FREUDENTHAL

Mathematisch lnstituut iier Rijksuniversiteit te Utrecht Utrecht, The Netherlands H. de VRlES

Mathetnutisch Instituut der Katholieke Universiteit te Nijmegen Nijmegen, The Netherlands

@

ACADEMIC PRESS

New York and London 1969

COPYRIGHT 01969, BY ACADEMIC PRESS,INC. ALL RIGHTS RESERVED NO PART OF THIS BOOK MAY BE REPRODUCED IN ANY FORM, BY PHOTOSTAT, MICROFILM, RETRIEVAL SYSTEM, OR ANY OTHER MEANS, WITHOUT WRI'ITEN PERMISSION FROM THE PUBLISHERS.

ACADEMIC PRESS, INC. 111 Fifth Avenue, New York, New York 10003

United Kingdom Edition published by ACADEMIC PRESS, INC. (LONDON) LTD. Berkeley Square House, London, W1X 6BA

LIBRARY OF CONGRESS CATALOG CARDNUMBER: 68-23479 AMS 1968 SUBJECT CLASSIFICATION 2250

PRINTED IN THE UNITED STATES OF AMERICA

CONTENTS

. .

. .

. .

. .

0. 0.0

Introduction Preface. .

0.1-7.

On Notation and Nomenclature 0.1. 0.2. 0.3. 0.4. 0.5. 0.6. 0.7.

1-5.

Logical Synibols . Set Theory Symbols . Topological Symbols. Algebraic Symbols . Analytic Symbols . Logograms . . Nomenclature . .

. . .

. . . .

. . .

. . .

. . .

. . .

.

.

. .

. .

. .

. .

. .

. . .

. . .

. . .

. . . .

. xvii . xvii . xviii . xix

. . .

. . . .

. .

.

xx

xx xx

Preliminaries 1. Complex Extension, Real Restriction, and Waiving 2. The Exponential , . . 2.7-10. Examples . 3. Some Lie Groups . . . . . 4. Topological Groups . . . 4.1-7. Group Topology . 4.8. Coset Spaces . . . . . . 4.9-1 1. Local Groups . . . 4.12-13. Expanding Local Groups 4.1415. Local Coset Spaces . . . 4.16-17. Locally Connected Sets . . 5 . Differentiable Mappings

.

.

.

.

.

.

.

6-1 2.

. . . .

xv xv

.

.

.

. .

. . .

. . .

.

.

.

. .

. .

. .

.

. .

.

.

.

1 3 7 10 11 11 13 15 16 18 19 20

The Connection between Local Linear Lie Groups and Lie Algebras

. .

6. Definition of Local and Global Linear Lie Groups. 6.5. Historical Note . . . 7. The Infinitesimal Algebra of a Local Linear Lie Group 7.6. Examples . . 7.8. Historical Note . . 8. The Exponential Presentation . . . . . . 9. Homomorphisms, Automorphisms, and Derivations . . 9.2-4. Derivations . . . . . 9.5-7. Inner Automorphisms . . . . . . 9.8-1 I . Inner Derivations and Automorphisms . 9.12-13. The Topology of Int G . 10. Expanding Linear Lie Algebras and Their HomomcJrphisms into Linear Lie Groups and Their Local Homomorphisms . . . 10.7. Historical Note . .

.

.

.

V

.

.

23 24 24 29 31 31 35 36 37 39 41 42 46

vi

CONTENTS

.

11. Dropping Differentiability Assumptions . . 11.1-3. Closed Local Groups . . . 1 1.4. Continuous Homomorphisms . 11.7. C2-ConnectedSubgroups . . . 11.8. Historical Note . . . 12. Subgroups and Subalgebras, Normal Subgroups and ldeals 12.1-3. Closed Local Subgroups. . 12.4. Normal Subgroups and Ideals . 12.5-6. Abelian Subgroups and Algebras . . 12.7. Direct Products and Sums . . 12.8-9. Commutator Group and Ideal. . . . 12.10. Simple Groups and Ideals. . . 12.1 1-12. Coset Spaces . . . 12.13-14. Factor and Mod Reduction . . 12.15. Centerfree Groups . . . . . . I 2.16. ad-Closed Groups . . . 12.17. Historical Note .

. .

. .

.

13-19.

. .

.

.

Solvability and Semisimplicity

.

Solvable Groups and Solvable Lie Algebras Invariants of Linear Lie Groups and Algebras Rootsand Rank . . . . Important Classes of Complex Lie Algebras . 16.2. G E A! . . . 16.3. G E D1 . . . 16.4. G E BI . . . 16.5. G E C , . . . . . 17. Solvable Subalgebras . 17.1-3. Nilpotency 17.4-12. Trunks . . . . 17.13-15. Rootforms and Nodes. . 17.16-17. The Killing Form and Solvability 17.18. The Real Aspect . . . 17.19. Historical Note . . . 18. Cleaving . . . . . . 18.7. Historical Note . . . 19. Semisimplicity . . . 19.1-13. Criteria . . . 19.14-16. Irreducible Algebras . . 19.17. TheReal Aspect . . 19.18-22. Purity and Cleaving . 19.23. Historical Note . . .

13. 14. 15. 16.

.

.

.

.

. .

. .

. .

20-27.

. .

. .

. . . .

.

.

. .

.

.

.

.

.

.

.

. . .

. . .

. .

. .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

. . .

.

. .

. .

. .

.

. . . .

.

.

.

.

.

.

.

. . .

. . .

.

.

.

.

.

.

. .

. . . . .

.

.

.

. . .

. . .

.

.

. . . . . .

. . . . .

46 46 50 51 52 52 52 53 54 55 56 58 58 59 60 61 62

63 70 73 75 76 78 80 80 81 81 83 86 87 88 88 88 92 92 92 95 96 96 98

Dressings and Classification o f Semisimple Complex Lie Algebras 20. The First Dressing of Complex Semisimple Lie Algebras 20.14. The Trunk . . 20.5. Ladders . . . . . . . 20.67. Branches * . . 20.8-12. Ladders of Rootforms . . . . 20.13. First Dressing . . . 20.14. Historical Note . 20.15-16. Semisimple Subaigebras . . .

. .

. .

. .

.

. . . . .

.

. .

.

.

. .

.

. .

99 99 100 103 103 105 106 106

vii

CONTENTS

21. The First Weyl Norming and the Second Dressing of Complex Semisimple Lie Algebras . . 21.1. Norming the Branches . 21.2. The Standard Trunk . . . . 21.3. The Reflections . . . 21.4. Second Dressing . . . 21.5. Gordon Brown's Formula . . . . . . 21.6. Historical Note . . . . 22. GDetermined by W * . . . . . . 23. The Second Weyl Norming and the Third Dressing of Complex Semisimple Lie Algebras . . . . 24. The Unitary and Standard Restrictions of a Semisimple Lie Algebra 24.1-4. Standard and Unitary Semimorphism . . 24.5. Historical Note . . . . . . 24.6-7. Examples . . . 25. G Determined by W" . . . . . . 25.1. Angles of Rootforms . . . . . 25.2. Natural Bases and Ordered Dressing . . 25.3. TheGraph . . . . . . 25.4. Building W +from W++ . 25.5. Direct Splittings . . . . 25.6. The Top Rootform . . . 25.7-8. Examples . . . . 26. Classification of SemisimpleComplex Lie Algebras Up to Isomorphism 26.24. Complex Classification Theorem . . 26.25. Historical Note . . . . . 27. G2and F,. The Chevalley Dressing . . . . . 27.1-2. G2 and F4 . . . . . . . . . 27.3. The Chevalley Dressing . . . . 27.3.4. Historical Note . .

.

. . . .

.

.

.

.

.

. . . . .

. .

.

.

28-38.

. .

. .

. .

. .

.

.

.

. . . .

. . .

. .

.

. .

. .

.

.

.

. .

.

.

. . . .

. . . .

.

.

. . . . .

. . . . . .

.

.

107 107 109 109 110 111 111 111

115 117 118 119 120 122 122 123 125 126 127 128

128 131 136 137 137 137 139 141

Topological and Integration Methods

.

.

.

28. Homotopy and Wrapping . . . . . 29. Fundamental Groups and Wrappings of Topological Groups 30. Compactness Aspects of Semisimple and Abelian Groups . 30.4. Compact Inner Classes . . . . 30.5. The Global Structure of Abelian Linear Lie Groups . 30.8. Historical Note . . . 31. The Conjugacy Theorem for Centerfree Unitarily Restricted SemisimpleLieGroups . . . . . . . . . 31.1 3. Historical Note . . 32. The Fundamental Group of Centerfree Unitarily Restricted SemisimpleGroups . . . . . . . 32.12. Historical Note . . . . . 33. The Automorphisms of Semisimple Lie Groups . . . 33.1. Reflections as Automorphisms . . . . 33.2-3. The Automorphismsof W* . . . . 33.4-5. Chambers . . . . . . . . 33.6-8. An Alternative Approach . . . . 33.9. Automorphisms of G and G . . . . 33.10. Automorphisms and the Fundamental Group . . . 33.1 1. Plus-Automorphism . . . . . . . 33.12-13. Minus-Automorphism. . . . . . . 33.14. The Corner Lattice . . . . .

.

.

.

.

. .

.

.

.

.

.

.

.

.

.

. .

.

.

. . .

.

.

.

. . . .

.

.

.

.

142 150

152 152 153 154 154 160

160 166 166 167 168 170 171 173 175 175 176 177

...

CONTENTS

Vlll

34.

35.

36.

37.

38.

. . .

33.16. Kaleidoscope Groups . . . 33.17. Historical Note . . . . . . Integration in Compact Groups . . . 34.1-2. Measure in Groups . . . . 34.3. Measure in Compact Groups . 34.4. Historical Note . . . . . The Conducibility Theorem . . . . 35.1-2. An Invariant Inner Product . . . 35.3-4. Conducibility . 35.7-9. Compact Groups . . 35.10. Historical Note . . , Orthogonality Relations . 36.1-6. Irreducibility and Equivalence . 36.7-9. The Universal Representation and Its Components 36.10-1 1. Orthonormality . . 36.12-15. Totality . . . . The Characters of Compact Groups . 37.1. Orthonormality . . . . . 37.24. Reduction to the Trunk . . . 37.5. The Natural Gauge . . . . 37.6. Historical Note , . . * . . Some Global Properties of Semisimple Linear Lie Groups 38.14. The Hermitean-Unitary Split . . . . 38.5. Fundamental Group, Center, Closedness . . 38.6. Historical Note . .

.

. . .

. . . .

. . .

. .

.

39-50.

.

. . .

. . . . .

.

.

. . .

.

.

.

.

. .

. . . ,

. .

. . . . . . . . . . .

.

178 179 179 179 180 . 183 . 184 184 . 184 . 186 187 187 . 187 . 189 . 190 . 191 195 . 195 . 196 199 . 203 . 204 . 204 . 205 . 206

. . . . .

. . . .

The Algebraic Approach t o Linear Representations 39. The Associative Envelope of a Lie Algebra . . . . . 39.8. Historical Note . 40. The Casimir Tool . . . . . . 40.4. Historical Note . * . . . . . . 41. Weights and Integral Forms . * . . . 41.1. Weights , . . . . . 41.2. Ladders . . , 41.3. Ladder Length . . . . . . 41 4-5. Integral Elements . . . . 41.6-7. The Weyl Tool . . . . . 42. Source, Top Weight, and Limitation of a Representation . 42.1. Source . . . , . 42.2-3. Source of Top Weight . . . . 42.4. An Inequality . . . . . 42.5. Finite Dimensionality and Irreducibility . . . . 43. Finite-Dimensional Irreducible Representations . . 43.2. An Alternative Approach . . . . . . . 43.3-6. The Value of the Casimir Tool . . 44. The Construction of All Finite-Dimensional Representations . 45. The Fundamental Weights . . . . . . . . 45.6. Integral Forms, Dominant Elements, and Fundamental Weights and Representations . . . . . . . 45.8. Historical Notes to Sections 41-45 . . . 46. The Fundamental Group of Unitarily Restricted Semisimple Lie Groups . . . .

.

.

. .

.

.

.

. .

.

. .

.

. .

. . . .

.

.

. . .

.

. .

. .

.

.

. . . .

.

.

.

.

.

.

.

.

.

.

. .

.

.

.

. . .

.

207 210 21 I 21 3 213 213 214 215 21 6 216 217 217 218 219 220 22 1 223 223 225 229 230 233 234

ix

CONTENTS

47.

48. 49.

46.14. Reenvisaged . . . 234 46.5-6. Fundamental Group of a Representation . . 236 46.7-9. Wrappings Realized . . 231 46.10. Another Approach . . . . . . 238 Weyl’s Character and Dimension Formula . . 238 47.1-6. Skew Functions and Characters . . 238 47.7. Weyl’s Character Formula . . . . . 242 47.8. Weyl’s Dimension Formula . . 242 47.9. Historical Note . . 242 47.10. An Elaboration . . . . 243 47.1 I . Strange Formula . . . 243 Algebraic Proof of Weyl’s Formulas . . 243 . 247 48.8. Historical Note . . Clifford Algebras and Spin Representations . . . 248 . 248 49.1-3. The Infinitesimal Approach . . 250 49.4-6. The Group Approach . . . 49.7. Reduction of Irrationalities . . . 252 49.8. The Other Fundamental Representations . . 255 255 49.9. Historical Note . . . The Conducibility Theorem (Algebraically Proved) and E. E. Levi’s 256 Theorem . . . . . . . . 256 50.1. Conducibility Theorem . . . . 256 50.2. E. E. Levi’s Theorem . . 264 50.12. Historical Note . .

.

-

50.

51-62.

. . .

.

.

Reality in Lie Groups and Algebras and Their Linear Representations

. . .

51. Maximally Compact Dressing . . . . 51.1-3. Preliminaries. . . . . 51.4. The Class of CC.. . . . . 51 S.1-10. C-Third Dressing and the Hermitean-Unitary Split. 51.5.11-12. The Maximal Compact Subgroup 51.6. C-Third Redressing . . . . . . . 51.7. Kinds of Trunks . . . . . . 51.8. Maximally Compact Trunks . . . 51.9. Criteria on Maximally Compact Trunks . 51.10. Conjugacy of Maximally Compact Trunks . 51.1 1. Maximally Compact Dressing . . . . 51.12-15. Reductions toward Classification . . 51.16. The Twin Case . . . . 5 1.17. Classification Dressing . . . . 51.18-20. Isomorphisms . . . . . . 51.21. The Description of the Graph of Cc,,, . . 51.22. Further Notations and Conventions . . 51.23. Historical Note . . . . . 52. Classification of Inner Types . . . . . 53. Classification of Outer Types . . . 54. Further Remarks on Real Classification . 54.1. Examples . . . . . . . 54.2. Maximal Signatures, Standard and Near Standard 54.3. Central and Near Central . . . . . 54.4-5. Coincidences . . . . .

.

.

.

.

. . . .

.

. .

.

.

. . .

. .

.

. . .

265 265 266 , 267 269 269 271 271 . 272 . 273 . 273 . 274 . 278 . 278 279 280 . 280 . 281 281 . 287 . 291 . 291 . 293 . 294 . 295

. . . .

. .

.

C0NT ENT S

X

. 295 55. Contravalence and Virtual Reality of Linear Representations . 55.2. Contravalence . . . . . . . . . 295 . . . . . . 296 55.3. Extension and Restriction . . . . . . . . . 296 55.4. Virtual Reality . 55.5-8. Waiving and Twinning . . . . . . . 297 . . . . . . . 298 56. Contravalence of Weights . . . . . . . . . 300 57. Self-Contravalence . 57.1-7. Toward a Criterion on Virtual Reality . . . 300 57.8. Antireality and Quaternion Space . . . . . 306 57.9. Historical Note . . . . . . . .306 58. Computing= for Simple Lie Algebras . . . . . . 306 59. Invariant Bilinear and Sesquilinear Forms . . . . . 310 . . . . . . 310 59.1. Unitary Representations . 59.2. Uniqueness of Invariants . . . . . . . 311 . . . . . . . 312 59.3. EInvariant Forms . 59.4. GJnvariant Forms . . . . . . . . 313 . . . 314 59.5. Gc-Invariants in Real and Quaternion Space 59.6. Toward Computation 316 . . . . . . . .317 59.7. Examples 59.8. RealInvariants in the ArealCase . . . . . 319 . . . . . . . . 320 59.9. Historical Note 60. Minimally Compact Dressing . . . . . . . 320 . . . . . . . . 320 60.1-8. he-Notions . . . 323 60.9. Criterion on Trunks Containing a he-Trunk . . . 323 60.10. Gc,.,-Conjugacy of Trunks Containing a he-Trunk 60.1 3. Gc-Conjugacy of Trunks Containing a Maximal i-Compact 324 Subset 60.14. Real Rank . . . . . . . . . 325 60.15. A Procedure . . . . . . . . . 325 . . . . . 326 60.16-17. Minimally Compact Trunks. , , . . . . 326 60.18. Real Ordered Dressing . 60.19. How Far Is C Determined by Its Behavior on H . . . 327 60.21. A Procedure . . . . . . . . 329 60.22-23. Toward Computing Minimally Compact Trunks . . 331 60.24. Historical Note . . . . . . . . 334 61. Real Semisimple Linear Lie Groups as Products of Maximal Compact . . . . . . . . 335 and Solvable Groups . 61.5. Historical Note . . . . . . . . 337 62. The Fundamental Groups of the Real Types. . . . . 337 62.7. The Duals of the Fundamental Groups of Noncompact Center. . . . . 340 free Simple Groups of Inner Type 62.8. The Duals of the Fundamental Groups of Centerfree Simple Groups of Outer Type . . . . . . . 345 62.9. The Duals of the Fundamental Groups of the Universal Linear Wrappings . . . . . . . . . 347 62.10. Historical Note . . . . . . . .348

.

63-67.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

. .

Symmetric Spaces 63. Homogeneous Spaces and Riemannian Manifolds-A . . . 63.1-3. Homogeneous Spaces . 634 6 . Geodesics . . . . . . 63.7-9. Isometries . . . . . . 63.10. Historical Note . . . . .

Sketch

.

. . .

. .

349 349 . 352 .360 . 361

xi

CONTENTS

64. Symmetric Spaces . . . . . . . . 362 64.12. Historical Note . . . . . . 378 65. Minimal and Maximal Symmetric Spaces . . . . 379 65.4. Maximal Compact Subgroups . . . . 381 65.6. Historical Note . , . . . . . .382 66. Autometrisms of Symmetric Spaces, Automorphisms of Real Semi. . . 382 simple Lie Groups . 67. Fundamental Groups of Symmetric Spaces . . . 391

.

.

.

68-75.

.

Tits Geometries 68. A List of Fundamental Theorems . . . . . . 68.1. Trunks . . . . . . . . 68.2-3. Maximal Solvable Subalgebras . . . 68.4-6. Nonsemisimple Maximal Proper Subalgebras . . . 68.7. Parabolic Subalgebras , . . . 68.8. Nilpotents, and Semisimple Subalgebras of Rank 1 . 68.9. Bruhat's Lemma . . . * . . 68.10-1 1. Some Homogeneous Spaces. 68.12-14. A Proof . . . . . . . . . 68.15. Historical Note . . . . . 69. Proofs of the Statements of Section 68 . . . . . . 70. Introduction of Incidence Geometries of Semisimple Lie Groups . 70.20. The Special Case of rank 2 . . . . . 71. An Axiomatic Approach to Incidence Geometries of Semisimple Lie Groups . . . . . . . . . 72. Covariants of Pairs of Elements in Incidence Geometries of Semi. . . . . . . simpleGroups . . . . 73. The Classes of Pairs of Elements in an F4-Geometry . 74. The Incidence Geometries of Real Semisimple Lie Groups 75. C-Graphs of Incidence Geometries of Simple Semisimple Lie Groups 75.19. The Geometric Interpretation . . . . . . 75.20. Historical Note . . . . . . . .

.

.

.

.

.

.

.

76-77.

395 395 395 396 397 397 397 397 398 399 400 414 426 427 436 439 479 485 494 495

Betti Numbers of Semisimple Lie Groups and Regular Subalgebras of Semisimple Lie Algebras

.

76. ad-Nilpotents and Semisimple Subalgebras of Rank 1 76.10. Historical Note . . . . . . 77. Killing-Coxeter Tools, Betti Numbers . . . . 77.1. Introduction . . . . . . 77.2-6. Killing-Coxeter Tools . . 77.7-14. Killing-Coxeter Elements . . . 77.15-18. Int W* -Invariants . . . . . 7119-23. Int G-Invariants . . . . . . 77.24. RelatingB,.,(G) to the Center of d(G) . . 77.25-27. ,A Characterization of Killing-Coxeter Elements 77.28. Historical Note . . . . . .

.

.

.

. . . . .

. . . .

. . . . .

. .

. . .

.

497 501 501 501

505 512 518 521 523 524 525

xii

CONTENTS

Appendix TABLEA The Graph and Dimension of G E Alg Lie Com SSS;The Length Square of the Shortest Nonzero Rootform . . . 521

.

TABLE B The Positive Rootforms on a Natural Basis

.

.

528

TABLE C The Positive Rootforms on a Symmetric Basis and Their Altitudes

.

532

TABLE D Number of Rootforms of Given Positive Altitude a

.

.

.

.

534

TABLE E Dominant Rootforms TABLE F Fundamental Weights

.

.

.

.

.

.

.

.

.

.

.

.

535

.

.

.

.

.

.

.

536

TABLE G Isomorphisms and Equivalences for Low-Rank Semisimple Lie Algebras . . 538

.

Key t o Definitions Author Index

.

.

.

.

539 541

xiii

CONTENTS

Tables in the Text Z(fundanienta1 group) . . . . . . . . . Plus-Automorphisni . . . . . . M inus-Automorphism . . . . . . . . Kaleidoscope Groups . . Integral Forms, Dominant Elements, Fundamental Weights . Duals of the Fundamental Groups of Centerfree Gun . . . Classification of Inner Types Classification of Outer Types . . . . . Examples of Real Types . . . . Maximal Signatures, Standard and Near Standard . . . Self-Contravalent Weights . . . . E for Simple Lie Algebras . . . . . Examples of Invariant Forms . . Minimally Compact Trunks . . . . The Duals of the Fundamental Groups of Centerfree Simple Groups . The Duals of the Fundamental Groups of the Universal Linear Wrappings The Autometrism Classes of Symmetric Spaces. The Automorphism Classes of Simple Semisimple Real Lie Groups . . . . 67.2, 67.4 Fundamental Groups of Synunetric Spaces . . The Classes of Pairs of Elements in an F,-Geometry . 73 Minimally Compact Trunks (partly) . . . . . . 75.4 The C-Graphs of Incidence Geometries of Simple Semisimple Lie Groups 75.5-18 The Sequence k (for Betti Numbers) . . . . . 77.12

32.5 33.11 33.12 33.16 45.6 46.4 52 53 54. I 54.2 56.3 58 59.7 60.23 62.7-8 62.9 (56.7-8

. .

. .

.

. .

.

.

. .

.

.

.

.

.

. .

. .

.

165 i75 176 178 230 236 281 287 29 1 293 299 306 317 33 1 340 347 386 392 439 487 488 515

Graphs 25.7 26.23 27.1 27.2

Graphs of Lie Algebras Graphs of Lie Algebras G* F4

.

.

.

.

.

. . .

.

. . .

.

. . . .

. .

.

.

.

.

.

. .

. .

.

.

.

128 136 137 139

This Page Intentionally Left Blank

0. INTRODUCTION

0.0. Preface Be either consequent or inconsequent, never both together. (An unknown moralist.)

This is not a motto. It is rather an anti-motto. The authors have tried to be both. One may still wonder whether consequential inconsequence or inconsequential consequence prevails. Purity of method has been pursued, sometimes as an ideal, sometimes as a hobby, sometimes for no reason whatsoever. Impurity of method has been allowed for pragmatic reasons or because of its charm. Group and Lie algebra methods are by turns interwoven and neatly separated. Diction vacillates between formality and looseness. Function notation has been perfected, but still the authors struggled with derivatives. Categories have not been used, even where they were badly needed. The most serious flaw, however, is that the authors stuck to our grandfathers’ fields of the real and complex numbers. So neither algebraic groups nor Lie algebras over more general fields have been mentioned. In contrast, the restriction to linear Lie groups has been rather pragmatic. The present work is a textbook in the sense that it aims at leading the reader more and evermore quickly to attractive results, even if this means a maze of unrelated subjects in a seemingly illogical arrangement and a lack of lengthy introductions of a highly abstract character. On the other hand, it looks like a compendium where it goes into pedantic classificatory details, summarized in so many tables (those of Section 75 included) that before long it will belong to the class of books which contain all possible tables except those one needs at the precise moment. The reader of the present book is expected to have mastered the fundamental ideas of modern mathematics so that he may employ them with decent skill. The main subjects treated are classification of semisimple complex and real Lie algebras, their finite-dimensional linear representations and their automorphisms, the fundamental groups and Betti numbers of Lie groups, symmetric spaces, and Tits geometries. Some results and many methods are new. Deviations from traditional notations and nomenclature are justified after this xv

xvi

0.

INTRODUCTION

preface, and an inventory of definitions is found on page 539. References are always to chapters, sections, and subsections, never to pages. The present work has grown from gradually revised courses taught by the first author since 1931. From 1962 he has been assisted by the second author who has taken an ever increasing part in its contents and its formulation. Their cooperation has grown so close that finally the first of them does not understand how he could ever have thought to write this book alone. Utrecht, Nijmegen May 1967

HANSFREUDENTHAL H. DE VRIES

0.1-7. ON NOTATION AND NOMENCLATURE

0.1. Logical Symbols

1 (non), A (or), v (and), V, (there is an x),

--f

(if.. . then),* (if and only if), iff (if and only if).

A, (all x).

Remark These notations are didactically justified by their analogy with the set theory notation.

0.2. Set Theory Symbols

0(void set). Remark A void circle is the most intuitive symbolization of a void set. The struck out circle (or digit zero), quite usual nowadays, rather conjures the idea of a nonvoid set. This is clearly shown by the great confusion between = .@ and # n in the literature. During the first half-century of set theory, the void set was indicated by the same symbol as the number zero. Of course this had to be changed. A circle is a natural symbol for a void set and looks sufficiently different from a digit zero.

(element of), $ (not element of). (included in), 3 (includes), $ (negations of the preceding). u, U (union), n, n (intersection). (x E A \B) t t [(x E A ) A (x 4 B ) ] . E

+,

c

(x E { U I , . . ., a,)) tf v, (x = U J . Fa, b’ (the ordered pair of a, b). r a l , . . ., (the ordered n-tuple). r A , B I (the set of ru, b1 with a E A and b TA . . ., A,1 (analogously).

,,

Remark

E

B).

Brackets are needed for other purposes. It is desirable to spare

them.

txe F ( x ) A

(the set of x E A with the property F ) .

{x E AIF(x)) has several disadvantages: it spills braces, which are badly needed in punctuation and in pair forming, as tools for binding variables. It indicates the binding of a variable in a rather xvii Remark The usual notation

0.

xviii

INTRODUCTION

unusual way. It has been misinterpreted and misused so often that there is no longer a clear agreement on the place of the variable that undergoes binding. The presenc notation derives from B. Russell's i ( ( x E A ) A F(.Y)).To avoid typographical troubles, the roof has been put on a pole.

yx (. . . s . . .) yx, A

(. . . Y

(the expression between the parentheses 21s a function of s). . . .) (needs no further explanation).

Examples

yxs2

y

~

= square of . . . . log x = log.

If a group G is under discussion: Yxaxa-l = the inner automorphism of G induced by a. Y aY x a . ~ a -=' t h e canonical homomorphic mapping of G onto its group of

inner automorphisms. If on the group G a set of functions is considered :

2,F = Y a d a c ) g c Y Ya d a c )

-9 = yc y, y,rp(ac)

(right translate). (right translation). (representation by right translation).

Remark No doubt a symbol for function forming is badly needed. Traditional arrow notation like x + (. . . .Y . . .) would lead to clumsy expressions in the last few examples. The historical background of the present notation is analogous to that of Tx E A

F(.Y).

The composition of mappings has been denoted in the natural way rother than by a sign -1.

f l A (restriction of ;I mapping/'to ;I set A ) . = (equal), # (unequal). , =

c

Qi

X',

g ( X ) = C PI Xi,

h(X)=

E

End R

c yi xi,

if convergent. Then, obviously, h(A) = g(f(A)) as long as the underlying series converge in the norm.

2.

5

THE EXPONENTIAL

Define the logarithm by m

log(1 + A ) = c [(-l)"-'/n]A" I

as far as convergence in the norm prevails. Then by the foregoing exp log X = log exp X = X as long as convergence in the norm prevails.

2.3 In topological linear spaces it makes sense to speak of curves, tangent vectors, and so on. Definition A k-times continuously differentiable mapping of an open connected set of real or complex numbers into R is called a real or complex C k - c w r e on R. A C1-curve is simply called a curve. If x is a curve one defines

wherever this limit exists. (is)( t ) , also written

is called its rangerit uector at t, or at s ( t ) . Analogs of well-known rules are

and if A ( r ) , B ( t ) are continuous linear mappings of R into R, depending differentiably on r ,

2.3.1

d dt

d dr

x(r)+ A(t)-x(r),

d B(r) + A ( ? ) - B(r). dr If A ( t ) - ' exists, it is easily seen to be differentiable with respect to rand its derivative can be found by differentiating A ( t ) A ( t ) - ' = 1 :

2.3.2

thus,

2.3.3

d ( A ( t ) - I )= -A(r)-' dt ~~

1-5. PRELIMINARIES

6

2.4 It follows from the definition of exp that

d dt

2.4.1

- exp !A = A exp tA = (exp t A ) A .

For commuting continuous linear mappings A, B exp(A

+ B) = exp A .exp B.

This results from the same principle used at the end of 2.2. It is worthwhile, however, to prove it by a method that will be applied in the sequel in a more profound way: First 2.4.1 is generalized. 2.4.2 For AB = BA, (d/dt)etA+B = AefAtB Indeed,

d 4! ((tA + B)"' = m(tA + B)"'-' - ( t A + B) dt dt

because all things commute. Therefore,

d dt

l (tA + B)" m.

=A

2

m - I)!

which proves 2.4.2. Now d (e-tA etA+B) = -Ae-tA etAtB + e-tA dt

(tA + B)"'-l,

AetAtB =0

because factors commute. So C = e-fA

erAtB

is constant. Substituting t = 0, one gets

C = eB. Substituting t = 1 : 2.4.3

e-A eA+B= eB

and, more particularly, for B = 0 thus,

e-A

eA = l ;

e-A

= (&)-I

~

Multiplying 2.4.3 by eA,one finally gets 2.4.4

exp(A

+ B) = exp A.exp B

for commuting A, B.

2.

7

THE EXPONENTIAL

2.5 A consequence of 2.4.4 is

exp(T, + T J A = exp T , A . exp T~ A . This means that

Y f exp t A is a homomorphic mapping of the addition group of scalars. It is a continuous and even a differentiable homomorphism:

d - exp t A = A exp t A . dt 2.6t Any A ( t ) as a function o f t with A(0) = 1 may be interpreted as aflow in R : the particle that was in a at time 0 will be in A ( t ) a at time t . In this sense (d/dt)A(r)is the velocity field of the flow at time t . If the flow is of the special kind A ( t ) = exp t A , the particle that was in a at t = 0, and consequently in b = (exp rA)a at time t , will show there the velocity A(exp tA)a = A b at time t. This means that the velocity at any time depends on the spatial spot only, not on time. Particles passing through a given spot b will have the same velocity there. This kind of flow is called stationary. Y t exp tA yields a stationary flow with the velocity field

(i

A = -exptA )r=o

.

It is clear that in a stationary flow all particles passing through b will arrive at some other spot in the same lapse of time T so that A(t T ) * awill not depend on t, provided A ( t ) a = 6 . In other words, A ( t T ) A (t)-' does not depend on t , thus equals A ( T )which just restates the homomorphic character of yrA ( t ) .

+

+

2.7-1 0. Examples 2.7 Suppose that dim R = n < co. Assume an ordered basis in R. Then a linear mapping is described by a matrix. Let A be triangular with zeros below the main diagonal. Then Am is also triangular and so is exp A . To every diagonal element h of A corresponds eAof exp A . Thus, 2.7.1

det exp A

= exp

In particular,

t This section will not be used in the sequel.

tr A .

1-5. PRELIMINARIES

8

if tr A = 0, then detexp A = 1.

2.7.2

These formulas are generally valid because over the complex field and on a suitable basis, a linear mapping of R (dim R < m) takes the triangular form. 2.7.3

detexpA # 0 forall A.

2.8. Suppose

dim R = 2,

A

=

(: k).

Then

Generalization: let A be n-dimensional triangular with a vanishing main diagonal. Then in A 2 all elementsjust above the main diagonal also vanish and so on; finally, A” = 0. Thus exp T A is a polynomial in T of degree SG n - 1. 2.9 Suppose that A is skew. With A’ the transpose of A, A’ = - A ;

thus, (A”)’ = (-A)m,

(exp A)’ exp A

=

1.

Therefore, if A is skew, exp A is orthogonal. In the same way one proves the following. If A is hermitean skew, then exp A is unitary. (Hermitean skew means that A* =-A, where A* is the conjugate transpose of A ; unitary means that A*A = 1.) 2.10 t The importance of the exponential is illustrated in a quite informal way by the next example. Let R be the linear space of functions of a real variable in [-a,031. The substitution 5 + 5 t in the argument produces a mapping T, of R onto itself:

+

(7-2 9)(5)= dt + 0 ; T, transfers the graph of y over a distance t to the left; T, is linear, and, because of (TsTt 9 M )= 9(t + t + s) = V S + , d ( 5 ) Y

t This section is not used in the sequel.

2.

9

THE EXPONENTIAL

Y,T, maps the addition group of real numbers homomorphically. One might expect it to be a stationary flow. Differentiating, one gets

(;

T,).)

d

(0= g ( t + t ) = D d f + t ) = (Tt D d ( 0 ,

where D is the differentiation operator,

Thus, d -Tt dt

=

T , D = DT,.

D is the velocity field of the flow. (Note that it is restricted to the subspace of differentiable functions.) Analogy suggests that T, = exp tD,

which means that

that is, Taylor's formula. Because of its restricted validity, the analogy is merely formal. By a Fourier transformation of this example, another develops: (Zcp)(q)= (2n)-'I2

I" e-'c? ~ ( 5 )

df

-a

defines the Fourier transform Z v of v in a certain subspace of R . It is invertible there : ( Z - l v*)(f) = (2n)-'12 v*(q)d7. -a

s"

Translating T, from the "planguage" into the "q*-language," one gets

P,= ZT,Z-'. Now (z-1

P,v*)(f) = (T,Z - 1

v*)(o= ( 2 7 T ) - 1 1 2 J m--m = (2n)-'12

thus,

(P,v*)(7)= e'?'

V*(T).

eiq(e+t)v* ( 4 4 e'c? e'?' v*(q)dq,

1-5. PRELIMINARIES

10

Again, ps+t = pspt. y t P t maps the addition group of real numbers homomorphically. For the velocity field A of this flow one finds

A is the multiplication by i times the argument. Again formally, P, = eta.

3. SOME LIE GROUPS The groups that appeared in Section 2 were homomorphic images of the addition group of real or complex numbers. This section is a preliminary exploration of groups in which the general element depends on more parameters. 3.1 The multiplication group of n-n-matrices with determinant # 0 (or the group of automorphisms of linear n-space). The element a depends on n2 parameters, the matrix coefficients all (a).

3.2 The subgroup of 3.1 singled out by the condition det a = 1. The group element a now depends on n2 - 1 parameters, one matrix coefficient being redundant: a. has some matrix coefficient au(ao)with nonvanishing minor; near a. the equation det a = 1 can be solved with respect to cq,, and a is fully described by the remaining matrix coefficients as parameters. The validity of this parameter system is merely local; it breaks down as soon as the minor vanishes.

3.3 The subgroup of 3.2 consisting of orthogonal a ; thus a‘a = 1. Later on it will be shown that a depends on +n(n - 1) parameters. 3.4 The group of complex n-n-matrices a with a*a = 1 (unitary matrices). The number of (real) parameters is n2. 3.5 The group of matrices

with real a,/3 and /3> 0; two parameters.

3.6 The group of matrices

riise2it),

s, t real.

I1

4. TOPOLOGICAL GROUPS

There are two parameters s, t. The pairs rs,,tll, rs2,t21 produce the same element iff s2 - s, and t 2 - t l are integers. The topology of this group is that of the torus; a model is the square defined by 0 G s G I , 0 G t G I , with identification of opposite sides. 3.7 The subgroup of 3.6 defined by

s=

CLT,

t=/3~,

a, /3

fixed,

/3#0,

T

variable.

There is one parameter T . In the model of 3.6 the subgroup shows up as a straight line in the rs, t l-plane brought back piecewise into the square by reduction mod 1. Two cases are to be distinguished : 3.7.1 a//3 rational. Then CLT, / 3 are ~ both integers for some T-interval suffices to describe the subgroup.

T,

and a finite

3.7.2 a//3 irrational. Different T furnish different group elements. The group 3.7.2 is dense in 3.6. In 3.7.2 one can distinguish two topologies, one borrowed from 3.6 but pathological for 3.7.2 and the other that of the straight line, which is more adapted to 3.7.2. This phenomenon explains some precautions which will be taken in the fundamental definition of Section 6. 3.8 The group consisting of the (real or complex) upper triangular n-nmatrices (those of the form with 1’s in the main diagonal); the number of parameters is +n(n - 1).

4. TOPOLOGICAL GROUPS 4.1-7. Group Topology 4.1. Definition A topological group (Gru Top) is a set with a group structure and a topology such that multiplication and inversion are continuous operations in the given topology. It suffices to suppose that ab a-l ab

is a continuous function of ra,bl at ‘1,l l ; is a continuous function of a at 1;and is a continuous function of a as well as of b.

12

1-5.

PRELIMINARIES

)-l at x is proved by considering ~ - ~ ( a x - ~ for Then the continuity of fixed x and that of )' ra,biab at r ~ , y by l considering x(x-'a)(by-')y for fixed x, Y.

4.2 In groups, AB means the set of ab with a E A, b E B; A-' means the set of a-1 with a E A. 4.3 The continuity of group operations at 1 can be expressed as follows: for every 1-neighborhood U there is a 1-neighborhood V such that V-I c U, V V C u. 4.4 In a topological group, left (right) multiplication with a fixed element and inversion carry open sets into open sets. 4.5 In a topological group the 1-component is a closed normal subgroup. 4.6.1 If A c G E Gru Top, then the intersection of all UA with U ranging over the set of the 1-neighborhoods in G equals the closure A of A. Indeed x E A t t A" U - ' x n A # 0++ x E nu UA. 4.6.2 Pc YVfor any 1-neighborhood Vin G This follows from 4.6.1.

E Gru

Top.

4.6.3 Any T,-group is Hausdorff and even regular. Indeed, given a 1-neighborhood U,there is a I-neighborhood V such that V V c U, but then P= U. This property carries over to any point by left multiplication. 4.6.4 Any open subgroup of G E Gru Top is closed. This follows from 4.6.2 applied to an open subgroup V 4.6.5 A connected topological group (Gru Top Con) is generated by any nonvoid open subset. Indeed, the set generates an open subgroup which by 4.6.4 is closed; since the group is connected, it is identical with the whole group. 4.7 A discrete normal subgroup N of a connected topological group G lies in the center of G. Indeed, the set of axu-' (a E G) is connected for x E G, and discrete for x E N whence consisting of one point, which by taking a = 1 is identified as x. So axa-' = x for all a E G and x E N.

4.

TOPOLOGICAL GROUPS

13

4.8. Coset Spacest 4.8.1 lf H is a subgroup of the topological group G, the topology of G is transferred to the left coset space G / H by the convention that the open sets in G / H are just the images of open sets in G under the canonical mapping. The Hausdorff property and many others are preserved under this transition if H is closed. 4.8.2 Under the canonical mapping o f G onto G / H , the image of a connected set is connected; the inverse image of a connected set is connected if H is connected. In particular, if both G / H and H are connected, then G is connected. (Indeed, Y,aH maps open subsets of G onto open subsets of G / H even when restricted to some CH as a subspace of G.) 4.8.3. Definition A (continuous) representation .f of G E Gru Top in R €Spa Top is a homomorphic mapping of G into the group of autohomeomorphisms of R (notation: f = Yatc.fa) such that, in addition to the requirement that f a f b ' . f o b for a,b E G, Y r a . X i E r C . Rj ,fas iscontinuous.

The stabilitj>group o f sfl E R consists of the a E G with,fax, = xfl. For any A c G, fAxmeans the set of,f,x with a E A . ,fis called transitiz~eif for all x,y E R there is an a E G with,f,x = J'. 4.8.4 Clearly, if ,f is transitive, then the stability groups of all points are conjugate in C. If H is the stability group of so,then

maps G / H 0r.e-to-one and continuously into R. If/is transitive, this mapping is even onto. Proposition Let G be a locally compact Hausdorff group that satisfies the second countability axiom and let R be a locally compact Hausdorff space. Let G be transitively, continuously represented in R by f and let H be the stability group of .yO. Then

Y aHfabi

SO

is a homeomorphic mapping of G / H onto R.

t The results obtained i n this section will only incidentally be used in thesequel.

14

1-5. PRELIMINARIES

Proof It suffices to prove that there are “arbitrarily small” open I-neighborhoods U in G such thatfuxo is again open. Let a 1-neighborhood U3 in G be prescribed. Choose a compact l-neighborhood U in G such that u-’ U-’UU c u3.

Suppose thatfuxo has no interior. Thenf,,xo =f,fuxo has no interior either. Now every point of G is interior to some all (a E G);hence by second countability there is a sequence in G such that U, a, U = G.Put A, =f,,uxo. Then A, has no interior, U A , = R. A , is compact,

A decreasing sequence of V, c R will be defined such that V, compact,

V, has an interior,

V,,, n A , = 0:

V1 may be arbitrarily chosen to satisfy the first two requirements, and if V, has been determined then on the one hand V, has an interior, whereas A, has ‘\, , an interior point with a closed not, so, since A, is closed, one finds in ?A neighborhood V,+I also contained in V,\A,. Now n, V, has a void intersection with all A,, hence with R, whereas on the other hand it is nonvoid because of the compactness of the V,. This contradiction shows thatfuxo has an interior. Take an open U , with U c U o c UU. Then fuoxohas a nonvoid interior W.

UI = Uo n TaEG(f.xOE W) is open, and

xo= W is open. Finally, U2= U;’ U1 is an open 1-neighborhood in G,contained in U,-l U,, thus in the prescribed U 3 ,and fulxo =fu,-,fu,xo=fu,-l Wis still open. So U2 is the required “arbitrarily small” open 1-neighborhood with open fu2 xo. ful

4.8.5. Note More general theorems are found in H. Freudenthal, Ann. Math. 37,46-56 (1936). 4.8.6. Corollary Under the conditions in 4.8.4, if both R and H are connected, then G is connected. (See 4.8.2.) 4.8.7. Corollary Let rp be a (continuous) homomorphism of the locally compact Hausdorff group G with second countability axiom onto the locally compact Hausdorff group G’,with H its kernel. Then G / H and G‘ are topologically isomorphic in a natural way. Indeed, puttingfa = YxEGt(rpa)x for a E G,one gets a transitive representation f of G in G‘ with H as the stability group at the unit element of G’ to which 4.8.4 applies.

4. TOPOLOGICAL GROUPS

15

4.9-11. Local Groups

Let G

E

Gru Top.

4.9. Definition Two subsets of G are called locally identical if they coincide i n some I-neighborhood i n G (thus i n every sufficiently small one).

This notion will sometimes also be applied to subsets of differenr topoIogicul groups. The next definition is concerned with local subgroups of G. A tentative definition would read : H is a local subgroup of G if it is locally identical with H-' and with HH in a nontrivial way; that is, there is a 1-neighborhood U in G with 1 E H n U = H-I n U = HH n U.

Clearly nothing is lost if the requirement U = U-' is added. Further, if H is replaced by its local equivalent H, = H n U , then H,=H;'cU,

HIHlnU=Hl.

This consideration leads to a more practical definition : 4.10. Definition H is called a local subgroup of G if:

1 E H=H-'; H is contained in some open U = U - ' , such that HHnU=H. Moreover, such an H is called a closed local subgroup of G if it is closed in U. (It will be seen that the condition of closedness does not depend on the choice of U.) A great many properties of topological groups extend to local subgroups H . For instance, the group operations as far as defined in H are continuous. There is a 1-neighborhood Ho in H with H G 1 = Ho, HoHoc H. For any a E H there is a I-neighborhood H, in H with aH, c H. If P is open in H and aP(Pa) = H, then aP(Pa) is open in H, and so on. 4.1 1. Proposition A subset H of G is a closed local subgroup of G if and only if 1 E H = H-' and His open in the closure %of H H .

If: Let H be open in E.Then U = G\(%?\H) is open; if 1 E H , U is a I-neighborhood, and, if H = H - ' , then U = U - I . In any case, H c U , and H H n U = H, which shows that H is a closed local subgroup as soon as all conditions on H a r e satisfied.

Proof

16

1-5.

PRELIMINARIES

Only if: Let U = U-I be open, 1 E H, H = H-' closed in U, and HH n U = H. Then, with A = GU \,

HH= H u (HH n A),

-

H H ~ RH U H d u A;

intersecting with U,

H H n UcRn U = H , which proves that H i s open in E. 4.12-1 3. Expanding Local Groups 4.12 The local subgroup H of G will now be expanded into a topological group f i i n a natural and unique way. As a group, H will be the subgroup of G generated by H. The topology of however, is required to extend that of H in the following sense.

a,

Definition The topology of I? is said to extend that of H if H is open in I? and H i s a subspace of A; in other words, if the following applies:

4.12.1 Every set open in H i s open in I?. 4.12.2 Every set open in l? intersects H in an open set.

Suppose that such a topology exists. Then by 4.12.1 every aP ( a E I?, P open in H ) is open in Hand so is every union of such sets. C a l l 9 the set of unions of aP (a E A, P open in H). Then any member of .T is open in I?. Conversely, let Q be any open set in Z?, and c any point of Q. Then c-' Q is open in Hand, by 4.12.2, Q' = c-' Q n H open in H. So cQ' belongs to Y, and, since 1 E Q', c E cQ' c Q, there is a member cQ' of 9containing c and contained in Q. This shows Q as a union of members of 9, and therefore Q E 9. Thus, if it exists, the topology of I?is described by9- as its set of open sets. Its existence is guaranteed as soon as the intersection of any two members of 9(defined as above) belongs to 9. Set theory distributivity allows restriction to the case of aP and a'P', where a,a' E I?, P, P' open in H.Without loss of generality, one may even suppose that a' = 1 and P' = H. Let

c E a P n H. One then has to find a neighborhood Q of c in aP n H. One knows CEH

and

a - ' c E P c H.

4. TOPOLOGICAL GROUPS

17

Because of the continuity of left multiplication with a-' in H , one gets a c-neighborhood Q in H with a-' Q still c P ; hence Q c a p . So c E Q c UP n H , which proves the existence of the required topology of H. With this topology H appears to be a topological group: First, multiplication and inversion are continuous at 1, because they are so in H. Left multiplication is continuous because it leaves Y invariant. The continuity of right multiplication requires Pb to be open for any open P in fi and b E H . To prove this one may assume that b E H (which generates I?) and furthermore that P is open in H. Given c E Pb, c has to be proved an interior point of Pb. Now b-' E c-IP n H , which is open in H ; thus there is a 1neighborhood Q in H such that Qb-' c c-'P n H ; hence c E cQ c Pb, which shows the continuity of right multiplication. This proves the following proposition. Proposition If H is a local subgroup of G, then A, generated by H and provided with the unique topology that extends that of H , is a topological group. If H is connected, then fi is also connected. 4.1 3 Note that the topology of H may differ from that induced by G. This is illustrated by 3.7.2, in which 171 iw defines a local subgroup H of the topo-

logical group G of 3.6. Its H i s essentially the addition group of real numbers with the ordinary topology. As a dense subgroup of G, however, it bears another topology. Yet the following is true. Proposition If H is a closed local subgroup of G and fi bears the topology of subspace of G, then H has to be closed. If G is locally compact and satisfies the second axiom of countability, the converse still holds: if I? is closed as a subset of G , it bears the topology of subspace of G.

Let fi bear the topology of subspace of G. Since H i s open in H , one finds a U open in G with H n U = H and H closed in U. Let a belong to the closure of fi. There is then some b E H n aU. Now a-lb belongs to U and to the closure of H , so that it belongs even to the closure in U of I? n U = H, hence to H (and A ) , since H is closed in U . From a-l b E fi and b E H it follows that a E H. Consequently, l? is closed. Conversely, if G is locally compact, the closed subspace H is also locally compact, and 4.8.7 applies; namely, the identity mapping of into G is homeomorphic and therefore the topology of Hcoincides with that induced by G in I?. Proof

Remark There is no real need of the Hausdorff property, for the assertions

are not influenced by the factoring out of the closure of { l} in G.

18

1-5.

PRELIMINARIES

4.14-15. Local Coset Spacest 4.14. Definition If G E Gru Top and H is a closed local subgroup of G, then for any sufficiently small 1-neighborhood V = V-I in G the local c o w space V/His defined as the set of the a H n V with a E V, topologized by calling the subsets of V / H open whose unions are open in V . To justify this definition, one has to find a Vsuch that the sets aH n V constitute a partition of V :

[ a ~ V / ( b ~ ( a VH) ]n- + [ a H n V = b H n V ] .

4.14.1

This is achieved as follows. As in 4.10, take an open 1-neighborhood U

= U-l

in G such that

closed in U. HH n U = H Take a 1-neighborhood H , in H such that

HI HI =H and a 1-neighborhood V = V-' in G such that VVcU and H n VVcHI. To verify 4.14.1 for a E V , take

(i=l,2).

b,EaHn V Then

a-' 6, E H n a-' V c H n V V c H , ;

hence

by1b2 = (u-' bl)-' (a-' b,)

E

HI HI c H ,

and likewise

which proves 4.14.1. Proposition If G E Gru Top and H and H ' are closed local subgroups of G which coincide locally and such that H c H ' , then for sufficiently small 1-neighborhoods V both V / H and V / H ' make sense and coincide. Moreover, if H is a closed subgroup of G , then for open or compact V , V / H is topologically a subspace of G / H .

t The results of 4.14-15 will be used only incidentally.

4.

TOPOLOGICAL GROUPS

19

Proof Clearly V can be chosen so that V/H as well as V/H' are well defined, so one has only to make sure that (aH' n V = bH' n V ) + (aH n V = bH n V ) , in other words, b E a H ' n V + b E aH n V ; but this is evident as soon as V is so small that H and H' coincide in VV.

The remainder of the proposition is evident. 4.1 5. Proposition As defined in 4.14, V/H is Hausdorff and even regular. Proof With U and Vas above and a E V , His closed in U, so that H n a-l Vis closed in U n a - l V . Now a - ' V c V V c U , U n a - ' V = a - l V ; hence H n a-I V is closed in a-I V, aH n V is closed in V , which proves V / H to be a

TI-space. A neighborhood of aH n V i n V / H can be given in such a form that its union is Wa n V, where W is a 1-neighborhood in G . Let another neighborhood of aH n V have as its union W,a n V, where Wl is now a 1-neighborhood in G with W;' W , c W. To ensure regularity the union of the closure of the latter neighborhood in V / H must be shown to be contained in Wa n V. Indeed, let bH n V belong to this closure, with b E V. Then W ,bH n V, representing a neighborhood of bH n V in V / H , has a nonvoid intersection with Wl a n V. This shows that bh E W;' Wl a c Wa for some h E H , whence b H n V c W a n V. 4.16-1 7. Locally Connected Sets 4.16. Proposition Let G be a topological group satisfying the second

countability axiom, H a closed local subgroup of G, and fi defined as in 4.12. Let the subset A of fi be locally connected in the topology of G . Then the topologies induced on A by G and by fi coincide. Proof For given a E A it suffices to find a neighborhood W of a in A according to the topology of G such that on W both topologies coincide. As a matter of convenience one may suppose that a = 1. As in 4.14, Vis constructed on the evidence of H . Moreover, Vis assumed to be open and V / H is defined as in 4.14. Note that V, H, and V / H fulfill the second countability axiom. It is easily verified that the same is true of I?. 4.1 6.1

Y

C

W

n V)

maps A n V continuously onto a subset A' of V/H. In the fi-topology the different cH n V = c(H n c-l V ) with c E A n V are open and pairwise disjoint (see 4.15). Therefore, since I? fulfills the second countability axiom, their number is countable. Consequently, A' is countable.

20

1-5. PRELIMINARIES

Since V / H is regular (see 4.15), it may be presumed to be equipped with a metric, Its restriction to A' assumes a countable number of values. From this it follows easily that A' is of dimension 0.Therefore the components of A' are single points. Since A is locally connected, A n Vcontains a connected 1-neighborhood W (in the sense of G). Its image according to 4.16.1 is connected, thus a component of A', and thus a point, which must be H n I/. This shows W c H n V c H ; but by the construction of I? its topology coincides with that of G on H, hence on W. This proves the assertion. 4.17 The foregoing proposition will be applied later to make sure that a curve of G lying on f i i s also a curve on A.

5. DIFFERENTIABLE MAPPINGS

E, F E Spa Lin Top, dim E, dim F < 03. 5.1. Definition Iffmaps an open part of E into Fandfadmits of a linear mapping A of E into F such that f(X') -f(X)

= A(x'

-

x)

+ Ix' - XI E(X -+

Xo,

x'

--f

XCJ

(i.e., E goes to 0 if x and x' go to xo), then A is called the gradient mapping or the gradient off at xo, A = grad,,$ If it exists, A is unique. It exists as soon as f possesses continuous partial derivatives at x0 on some basis. A E is called the tangent space of the mapping at xo or, if confusion is unlikely, the tangent space at f ( x o ) . f i s called of class C k ( C m ;C m )if it possesses continuous kth-order derivatives (if it possesses derivatives of any order; if it is analytic). In statements involving C kit is understood that 03 and an are values of k. The inequality k < < an is assumed. In the case of complex E, Fit is known that C k = C""fork > 1. In this case by means of a semilinear K,one defines semi-Ck = K.Ck. Q)

5.2 Ifgfmakes sense and iffhas the gradient A at xo,and g has the gradient Batf(xo), then gfhas the gradient BA at xo. Iff-' makes sense, and iffandf-' have gradients, A , B at xo,f(xo),respectively, then B = A - * . If dim E = dim F = rank grad,,I, then f maps a neighborhood of xo onto a neighborhood of f ( x o ) , and then f-' exists locally near f(xo), and grad,,x,,f-' = (grad,,f)-'. If dim E = rank grad,,f, thenfis one-to-one near xo.

5.

DIFFERENTIABLE MAPPINGS

21

5.3 Manifolds will mostly occur smoothly embedded in finite-dimensional linear spaces. The next definitions concern such manifolds. For a more abstract definition, see 5.6. Definition A C,k-piece in Fis the image M of an open ball Sin E, dim E = r, by means of a mappingj; which is (a) homeomorphic, (b) of class Ck,and (c) provided with a nondegenerate gradxffor all x E S or, equivalently, provided with an r-dimensional tangent space at f ( x ) for all x E S. The pair ‘ E , f l is called the presentation of the C,k-piece M . The piece M is endowed with the topology induced by F. The notion of C1-curve for I < k on M , the notion of a C’-mapping and semi-C ‘-mapping of M , the notion of gradient of such a mapping, are explained in terms of E by means of the presentation of M . (They do not depend on the choice of the presentation.) Note that the tangent space of M at p depends continuously on p . Near p any affine projection of M on its tangent space is a C k-mapping with an identity gradient. Definition A C,k-manfold M in F is a connected topological space with the

following Properties: (a) M is a subset of F, (b) M is the union of a countable number of C,k-pieces M,, such that (c) any M, is a subspace of M , and (d) any M, is open in M . If ‘E,f’ is the presentation of a C,k-piece contained in M andfmaps the open ball S i n E such that its center is mapped intop, thenfis called a localpresentation of M near p . The foregoing remarks on tangent spaces, curves, mappings, and gradients extend to M . Note that the topology of M need not coincide with that induced by F. Whether a piece or a manifold is called real or complex depends on the underlying field of E. 5.4. Definition An infinitesimal measure m on the real C,’-manifold

M

is an assignment of a nondegenerate r-linear skew functional m, to the tangent space at p, depending continuously on p E M . An infinitesimal measure m on M can be integrated into a measure either directly by lifting the infinitesimal measure from the tangent spaces into M by means of affine projection and forming Riemann sums or indirectly by choosing a parallelepiped P in E (a sequence of r vectors), such that mx ((gradxf)(P))

’0,

and putting P ( W = S,.,,m,((grad,f)(P))dv(x) where Y is the ordinary measure in E, gauged by v ( P ) = 1.

22

1-5.

PRELIMINARIES

The result does not depend on the presentation of M. This follows from a formula known in calculus as “transformation of multiple integrals.” 5.5. Definition G is called a C,k-groupif it is a topological group, its space is a C,k-manifold, and its group operations are Ck-mappings with respect to the C,k-structure of the manifold G. The examples in 3.1-8 are Con-groups. For Aut R (3.1) it follows directly from the analyticity of matrix multiplication and inversion; for the others, it is less obvious but it will formally be proved in 11.3.5. Under left and right multiplication and inversion, the tangent spaces of a C,’-group are mapped by nondegenerate mappings (see 5.1). 5.6t Sometimes more abstract notions of C,k-manifolds and -groups are needed. They are described by the following. Definition A C,k-manifold consists of a pathwise connected topological space Mand a set @ of mappings with the properties :

5.6.1 Any cp E@ is a homeomorphic mapping of an open ball of an rdimensional E E Spa Lin Top into M. 5.6.2 The images of some countable number of

E@

cover M.

5.6.3 For qi E@, q;’tpl, as far as it is defined, is a Ck-mapping with nowhere-degenerate gradient.

It can be useful to add the following assumption. 5.6.4 @ is maximal with respect to 5.6.1-3. C,k-mappings and local presentations of such manifolds are defined in an obvious way, (The cp E@ are local presentations of the defined manifold.) To define the tangent space of rM,@l at some p E M ( p = rpa, where tp is some local presentation near a and a belongs to a ball S) an auxiliary tangent space of S at a is introduced : The set of C’-curves x in S with x(0) = a , [ d / d ~ x ( ~=) c,] ~is=called ~ the auxiliary tangent vector uc of S at a. The uc form the auxiliary tangent space of S a t a, with that structure of linear space in which u becomes a linear mapping. The pimage of an auxiliary tangent vector uc is the set of C’-curves 8 on rM,@l such that 8(0) = p and 8 coincides with some q x (x E oc) in a O-neighborhood; quc is called a tangent vector of rM,@l at p; these tangent vectors form the tangent space at p with that structure of linear space in which q becomes a linear mapping. The union over p of the tangent spaces at p E M of rM,@lis gifted with a structure of CrF-manifoldin an obvious way. ?The definitions of this section will only be used incidentally.

6-1 2 THE CONNECTION BETWEEN LOCAL LINEAR LIE GROUPS AND LIE ALGEBRAS

6. DEFINITION OF LOCAL AND GLOBAL LINEAR LIE GROUPS R

E Spa

Lin Top, E E Spa Lin Top, dim R = n < 03, dim E = r < 03.

6.1. Definition G is called an r-dimensional local linear Lie group (Gru Lie LOCLin) in R if it is a local subgroup of Aut R, and a C2-piece (in some ‘E, fl-presentation). G is called an r-dimensional (global) linear Lie group (Gru Lie Lin) in R if it is the extension according to 4.12 of an r-dimensional local linear Lie group in R. According to the field underlying E, a local G and its global extension are called real or complex. If R is real, then E is supposed to be real as well; if R is complex, then E is either real or complex. (Gru Lie LOCLin Rea or Com; Gru Lie Lin Rea or Com.) Real restriction and waiving in G are understood to be induced by the same operations in E if they again lead to (local) linear Lie groups. (Of course, a real restriction of the domain ball offmust again be a ball in the restriction of E.) Both are to be distinguished from real restriction and waiving in R (and consequently in Aut R), which may or may not accompany the corresponding operations in G . Complex extension of G, which must be some converse of real restriction, is explained in 10.6. 6.2 The examples in 3.1-8 will reveal themselves as linear Lie groups (see 11.3.5). Every linear Lie group will prove to be a Can-group (see 8.3). Meanwhile, a weaker assertion can be proved : 6.3. Proposition A linear Lie group is a C2-group. Proof Let G be the underlying local group; 6,according to 4.12, is a connected

topological group. The aG (a E 6) are C2-pieces because of the analyticity of left multiplication in Aut R and the nondegeneracy of the gradients. An everywhere dense countable subset A of G generates an everywhere dense countable subset B of The bG (b E B) form a countable system of C2-pieces covering 6 and are open subspaces of according to its construction. The

e.

e

23

24

&12. LOCAL LINEAR LIE GROUPS AND LIE ALGEBRAS

Cz-character of group operations in G follows from the analyticity of those in Aut R. 6.4 Incidentally it will be useful to consider a group GIthat is locally the isomorphic image of a given linear Lie group G. This means that by a certain tp a 1-neighborhood U in G is homeomorphically mapped onto a l-neighborhood U,in G, such that F(ab) = Fa-Fb as far as it is defined. Such a GI may possess several components, which, necessarily, are isolated. If such a GI is connected, it shares with G the same simply connected wrapping, as will be shown in Section 29. By means of the local presentations avf(u E G,,f as in 6.1) every component of G, is to be considered as a Cz-manifold [even a Can-manifold (see 6.2)], though not an embedded one. The 1-component then is a Cz-group (even a Can-group).

6.5. Historical Note Local Lie groups aredue to Sophus Lie (1842-1899). The adjective “linear” means the restriction to groups whose elements are linear mappings of some linear space. This restriction is pragmatic. An exact definition of general local Lie groups would be long-winded. The present methods are such that they can easily be adapted to general local Lie groups. Hilbert’s Fifth Problem asked for an elimination of differentiability assumptions from the definition of Lie groups. Its complete solution was reached in numerous steps from 1929 to 1952. (See D. Montgomery and L. Zippin, Topological Transformation Groups, Wiley (Interscience), New York, 1955; 3rd ed., 1965.) Its solution for local linear Lie groups is expounded in Section 11.

7. THE INFINITESIMALALGEBRA OF A LOCAL LINEAR LIE GROUP

G E Gru Lie LOCLin, dim G = r, G c Aut R, in ‘E,f ’-presentation. 7.1. Definition The tangent space of G at 1 is called G. Its elements are called the infinitesimalelements of G. Another way of looking at G will be useful: take a curve y, a, on G with a. = 1 and the tangent vector a=

(&)

. 1-0

G consists of all a. G is an r-dimensional linear subspace of End R. In 7.5 the following is proved.

7.

THE INFINITESIMAL ALGEBRA

Proposition a,b E G -+ab - ba

E

25

G.

7.2 The expression ab - ba is the keynote of the Lie theory. It is called the (infinitesimal) commutator of a, b. Definition For linear mappings a,b of a linear space into itself, one writes

[a,b]= ab - ba.

It is called the commutator of a and 6. [. .., ...] is anticommutative, [a,61

+ [b,a1 = 0,

and Jacobi-associative, “a,b],cl

+ “b, c1,aI + “c,al,bl = 0.

The first is obvious; the second results from simple computation. Furthermore, [. . .,. ..] is bilinear.

.

With [. . .,. .]interpreted as a product, G becomes an algebra, the injinitesimal algebra of G.

7.3 Disregarding the particular origin of the elements and the products [. ..,...], a general definition is given :

Definition An algebra with the product operation [. ..,. ..] is called a Lie algebra (Alg Lie, Rea or Com) if it fulfills

anticommutativity,

[a,b] +[b,a] = 0,

Jacobi-associativity,

[[a,b],c] + [[b,c],a]

and

+ [[c,a],b] = 0.

Complex extension, real restriction (if it leads to an algebra), and waiving in Lie algebras again lead to Lie algebras. From any associative algebra a Lie algebra is derived by putting [a,b] = ab - ba.

In fact, any finite-dimensional Lie algebra can be derived this way up to isomorphy, even as a linear Lie algebra as defined in 7.4. This is known as Ado’s theorem [see Trans. Amer. Math. SOC.Transl. 2 (1949)], which is not proved in this book.

26

6-12.

LOCAL LINEAR LIE GROUPS AND LIE ALGEBRAS

7.4. Definition A linear Lie algebra (Alg Lie Lin) in R is a Lie algebra contained in End R, with [. ..,. .] defined by [a,b]= ab - ba. R acted on by a complex linear Lie algebra G has to be complex; R acted on by a real linear Lie algebra G may be real or complex. Complex extension, real restriction, and waiving with respect to G are to be distinguished from the same kind of processes with respect to R, though in some applications they can go together.

.

Proposition 7.1 can now be stated as follows. Theorem The tangent space of G at 1 is an r-dimensional linear Lie algebra

G, the injinitesimalalgebra of G. Remarks (1) The term “infinitesimal algebra of G” is also used with global linear groups. (2) Infinitesimalelements and algebras, and subsets thereof, are

usually indicated by boldface type. If a (local) group is identified by some capital letter, its boldface counterpart usually indicates the infinitesimal algebra. (3) The converse of the theorem, that is, the unique existence of G E Gru Lie Lin with a given infinitesimal algebra G E Alg Lie Lin, is proved in Sections 8 and 10.1. It justifies the use of italic and boldface counterparts for related Lie groups and algebras. 7.5 Proof of Proposition 7.1. As a paradigm of future procedures, the fact that G is an r-dimensional linear space is restated in 7.5.1-3. 7.5.1

a

E

G +- aa E G

(a scalar).

Y t U r on G with a. = 1, a = [(d/dt)arlr,o. The curve Ytaarlies at least partly on G. Its tangent vector at t = 0 is [ ( d / d t ) ~ , , ]=, , ~ a[(d/dt)a,],,o = aa. Hence aa E G.

Proof There is a curve

7.5.2

a,bEG+a+bEG.

Proof There are curves )‘rat, Y,b, on G with a0 = bo = 1, a = [ ( d / d t ) ~ , ] , , ~ , b = [(d/dt)b,],,,.By c, = arbrfor small t , a curve on G is defined, and

which proves a

+b

E

G.

7.

THE INFINITESIMAL ALGEBRA

7.5.3

27

dim G = r.

Proof It may be assumed thatfO = 1 in the 'E,fl-presentation of G. On a basis of E curves on G can be expressed by at = f ' ~ , ( t ) , *,Tr(t)', with continuously differentiable p, and rp,(O) = 0. In particular,

v,(t)= t ,

for k # j

cpk(t)= O

defines a curve Y,a,(t) on G with

The a, are the images of the basis vectors in E under gradof, which is of rank r, and are linearly independent. Any a E G is the tangent vector at 1 of some curve defined by ar = f ' ~ i ( t ) , .,v X ~ ) ' ;

hence,

is linearly dependent on u l , ...,a,, which consequently form a basis of G. Remark Every a E G is a tangent vector of a C2-curve on G. Indeed, if u is the grad,jhiginal of u, then )',f(tu) has the required property.

7.5.4

a , b E G + [u,b]E G.

Proof According to the preceding remark, C2-curves Yrar, Yrb, can be found

with -ao = bo = I , .=(&a,)

r =o ,

b=(&b,)

The finite commutators for small t , c,

= a, b, a;'

define a curve Y, c, on G with co = 1. 7.5.4.1

hence, 7.5.4.2

b;',

t-0

.

28

6-12.

LOCAL LINEAR LIE GROUPS AND LIE ALGEBRAS

Also 7.5.4.3

Differentiating 7.5.4.1once more, putting t = 0, and using 7.5.4.2 one obtains

($a,) a;'

+ 2 (:a,)

+ a, ($a;')

($a;')

($a,),_o+ ($a;l)

7.5.4.4 7.5.4.5

= 0.

= 2a2. 1-0

( dt2 g b )1=0 + ( dt2 C b -'l )

=2b2.

Differentiating Yrc,, one gets

+ arbl(:a;')

6;'

+ arbla;' ( $ b ; l ) ,

which because of 7.5.4.2-3 shows

=o.

(ic,) 1-0

Differentiating once more, putting t = 0, and using 7.5.4.2-3,one gets

($

ct)t=o = 2ab - 2a2 - 2ab - 2ba - 2b2

+ 2ub + 2a2 + 2b2

= 2(ab - ba).

The statement 7.5.4.now follows from the following lemma. 7.5.5. Lemma If Yrc, is a @curve on G E Gru Lie LOCLin, and cg = 1, (dc/dt),,, = 0, then (d2c/dt2)1=o E G. Proof Y,c;'c,+,

is a C2-curve on G for any small r . Its tangent vector for

t=Ois c, = c;'

E G.

(;C+ l), 1=0

Since Cis a closed set, (d/dT)c,

E

G too, but

7. THE INFINITESIMAL ALGEBRA

29

7.6. Examples Though the Lie character of the groups G of 3.1-8 has not yet been formally discussed, it is possible to compute G : ad 3.1 Clearly G c End R. Let a E End R. If a, = exp tu then det a, # 0 and )',at is a curve on G with

hence G = End R. ad 3.2 If a, E G, det a, = 1, then d/dt det a, = 0. On an ordered basis a, is and det a, is a sum of terms &aI,, ... anJn. presented by a matrix r~i,(t)ly,,=l For t = 0, (xi, = 0, unless i = j . Every summand other than ccl .. a,, possesses at least two nondiagonal factors. Therefore after differentiation of det a, at t = 0 the only remaining contribution is

.

Thus u E G + tr u =O. Conversely, from 2.7.1, one learns that if tr u = 0 then det exp tu = 1. Then a, = exp tu defines a curve on G with [(d/dt)a,],,o= a. Consequently, u E G ++[a E End R A tr u = 01. ad 3.3 If a, E G, u = [(d/dt)a,],,o,then aia, = 1; hence

u'

+ u = 0.

Thus u is skew. Conversely, ifu is skew, a, =exp tu is orthogonal (see2.9)and [ ( d / d f ) ~ ,= ] ,u., ~ Consequently, G consists of the skew matrices of End R.

30

6-12.

LOCAL LINEAR LIE GROUPS A N D LIE ALGEBRAS

The skew matrices are also called infinitesimal rotations. The case of n particularly interesting. Then with

=3

is

one gets ax = [ u , x ] , where u = r ~ , , w 2w, 3 ' and [. ..,. ..] denotes the vector product; u is the axis of the infinitesimal rotation a and the rotations exp ta. (Note that au = 0.) ad 3.4 G consists of the hermitean skew matrices a of End R, a*

ad 3.5 G consists of the matrices

+ a = 0.

(: a):

ad 3.6 G consists of the matrices

(0" );

with imaginary u,p.

ad 3.7 G consists of the real multiples of

(Fi;)

ad 3.8 G consists of the triangular matrices of the form \ with zeros in the main diagonal. Thanks to the nilpotency of a - 1 for a E G, the power series of log a = log(1 + (a - 1)) converges (see 2.2). So log exists as the inverse of exp all over G and G, respectively; hence

exp maps G homeomorphically onto G. In all cases 3.1-8 one easily verifies directly that Cis a Lie algebra. In particular, 3.2 states that the commutator of two matrices with vanishing trace again hasvanishingtrace.ThisistrueevenofanypairA = ral,ly,,=l, B = rp,,l;,j=Lof n-n-matrices because tr AB = I; all p,l is symmetric in A , B. In 3.3 the case of n = 3 is again of particular interest. Given a , b E G, there are vectors u, v such that, for all x, ux = [ u , x ] ,

bx = [ v , x ] .

8.

31

THE EXPONENTIAL PRESENTATION

Then abx

=

[u, [u,

XI],

bax = [ v , [u,x]]. Computing the commutator of the matrices a,b, one finds

[a,blx = “ u , 01, XI ; thus, [u, [ V J I l

- [V,

[U,XIl =

“u,uI,xl,

which is a well-known property of the vector product, closely related to Jacobi-associativity. Lie algebras of linear mappings in function spaces play a role in quantum mechanics. Let Q,P, and I be defined by

Then [P,I ] = [Q, I] = 0.

[P,Ql The linear combinations aI

= I,

+ ,!?P+ y Q with scalar a,fl,y form a Lie algebra.

7.7 Real restriction and waiving in (local) linear Lie groups induce the same kind of processes in their infinitesimal algebras. 7.8. Historical Note Commutators and Jacobi-associativity appeared in Jacobi’s study of partial differential operators.

first

8. THE EXPONENTIAL PRESENTATION

G E Gru Lie Lin, G c Aut R, G locally presented as ‘E,fl, j-0 = 1, G its infinitesimal algebra. 8.1. Proposition The tangent space of G at a. is a. G.

32

6-12.

LOCAL LINEAR LIE GROUPS AND LIE ALGEBRAS

Proof By left multiplication with a,' a curve )',a, through a. on G is mapped into a curve Y,b, on G through 1,

The tangent space of G at a. is made up of the

which belong to aoG. 8.2 How does one reconstruct G from G? The examples 7.6 suggest that it could be done by the exponential. Indeed, local coincidence of G and exp G will be shown. The first step is the following proposition. Proposition a E G --jc exp fa E G.

Proof c, = exp fa as a function of t is characterized by the differential equation

d ;i;c, = c, a,

8.2.1

with the initial condition 8.2.2

co= 1.

The solution to 8.2.1-2 in End R is unique. If it can be shown that for small t 8.2.1-2 can already be solved on G, the uniqueness of the solution will guarantee exp fa E G for small t and so for all t because the exp fa form a group. According to 8.1,ca belongs to the tangent space of G at c E G. Marking the vector ca at any point c E G, one gets a continuous vector field on G. By integrating it? one obtains a curve y a, for small t, with a, E G, a. = 1, and such that for any t its tangent vector is just the prescribed a,a. Therefore the curve fulfills 8.2.1-2, which consequently can be solved on G. This proves the assertion.

,

t It is done by translating the problem into E throughf-I, where the resulting differential equation has to be solved.

8.

THE EXPONENTIAL PRESENTATION

33

8.3 The previous proposition shows: exp G c G. By the equality of dimensions and by the fact that grado exp is the identity mapping of G, it follows from Proposition 4.16 that exp maps a sufficiently small open 0-neighborhood of G homeomorphically and nondegenerately onto an open 1-neighborhood of G. This yields a new presentation of G near 1, E being replaced by G (as a linear space) and f by exp. The new presentation has the advantage of being intrinsic and analytic (because exp is so). By left multiplication this presentation is transferred to any point of G, which shows that G is analytic. This suggests the following definition. Definition The ‘E,f’-presentafion of G near 1 is called exponential if E = G (as a linear space) and f is the restriction of exp to an open ball around 0 in G . The following has been proved : Theorem A linear Lie group may be considered as an analytic group. Near 1

it admits an exponential presentation. 8.4. Definition An open ball N around 0 in G E Alg Lie Lin (according to some norm in G )is called a smooth ball if, in the closure of N, exp is one-toone and grad exp is nondegenerate. By exp the notion of smooth ball is carried to G, if G is the infinitesimal algebra of G. The existence of smooth balls follows from grado exp = 1 and the continuity of y. grad, exp. Note that if Nis a smooth ball then exp Nis a C”“-piece. Proposition If N is a smooth ball in G E Alg Lie Lin and N’ is a smaller concentric open ball in G such that

exp N .exp N’ = exp N, then exp N’ is a closed local subgroup of Aut R, as well as a local linear Lie group. This follows from 4.11. Proposition For the infinitesimal algebra G of the linear Lie group G, sets N, N ‘ as introduced in the preceding proposition exist and then exp N’ is locally identical with G.

8.5 The exp-image of a straight line through 0 as far as contained in N’ (see 8.4) is a one-dimensional local linear Lie group. This leads to the following theorem.

34

6-12.

LOCAL LINEAR LIE GROUPS A N D LIE ALGEBRAS

Theorem A 1-neighborhood in G is covered by a smooth system of onedimensional local linear Lie groups, intersecting in 1 only.

8.6 Globally things might be less smooth. The torus group 3.6 admits an infinity of one-dimensional subgroups through every one of its points. On the other hand, it may happen that G is not exhausted by exp G. As an example, take the group G of 2-2-matrices with complex coefficients and unit determinant. Then G consists of the 2-2-matrices I( with tr a = 0. There is no basis on which a=(-' 0 -1

I)~G

appears in diagonal form. The same can be said of a if exp a = a. Consequently, a should have two equal eigenvalues. Since tr a = 0, both must vanish and those of a should be 1, whereas in fact they are -1. This shows a 4 exp G. 8.7. Proposition Let K , ( i = I , . .. , k ) be C'-pieces in G , passing through I and with the respective tangent spaces Kf at 1. Let G = K f directly as linear spaces. Then there are C'-pieces K ; c K I with the same tangent spaces at 1 such that

z:

Y Tol, a i . . . .. 0 x 1 a , ' 0 2 ' . *

*

'ak

maps TK;,K;, . ..,KLl homeomorphically onto a I-neighborhood in G. I n particular, K , ' K , * * * - K k contains 1 in its interior with respect to G.

-

Proof The Kl may be assumed in 'E,,.f,'-presentations w i t h h t , defined for ti E SI and Sf open balls around 0 in E, ( i = 1,. . .,I?); the El may be assumed to be

direct summands of E = C El. Put

Thenfis a C1-mapping into G and (gradof)(z: II) = det(grad,f)

=

z: (gradoh) t l , det(grad,f,) # 0.

Sofmaps a small 0-neighborhood in E homeomorphically onto a I-neighborhood in G , which proves the assertions. 8.8 The special case in which Uis a small I-neighborhood in G and

K1 = (exp KI) n U, leads to the following.

9.

HOMOMORPHISMS, AUTOMORPHISMS, DERIVATIONS

35

Definition Let G = K , + K2 +** -+Kk directly as linear spaces and let w 1 be the natural projection of G onto K1. Then ‘E,fl is called a generalized exponential presentation of G near 1, induced by the aforementioned direct splitting, if E = G,

fa=expw,a.expw,a- * - . -exp wka. The existence of such presentations has just been proved.

9. HOMOMORPHISMS, AUTOMORPHISMS, AND DERIVATIONS 9.1. Local Homomorphisms G, H

E

Gru Lie Lin.

Definition 0 is called a local homomorphism of G into H if it maps a 1neighborhood in G continuously into H such that

O(a6)= @ a .Ob as far as @a,Ob, @(ab)are defined. Two local homomorphisms coinciding in some 1-neighborhood in G are considered identical. If such a 0 maps every sufficientlysmall l-neighborhood of G onto a 1-neighborhood in H , it is called a local epimorphism. The terms local endo-, iso-, and automorphism apply in an obvious way. It is also clear what is meant by a local C2-homomorphismand a local semi-C2-homomorphism (see 5.1, 5.3). A local C2-homomorphism 0 of G into H induces a linear mapping, also called 0 (instead of grad,@), of their infinitesimal algebras G into H such that if a=($a,)

Ec, t=O

then

lffor C2-curvesY,a,, Y,b,,

,

a = ($a,) t=O

then (see 7.5)

6=($b,)

EG, 1-0

36

6-12.

LOCAL LINEAR LIE GROUPS A N D LIE ALGEBRAS

[-

d2 (@a,* Ob, *(@a,)-' @b,)-l))] 2 dt2

=1

1=0

= [@a, 061.

This proves the following proposition. Proposition A local C2-homomorphism of linear Lie groups induces a homomorphism of their infinitesimal algebras.

In 10.4 the converse will be proved. In fact exp and 0 will be shown to commute. As a consequence, the validity of the prefixes epi, endo, iso, and auto, with respect to local groups and their infinitesimal algebras, will imply each other. It is clear how the proposition is to be stated for local semi-C2-homomorphisms.

9.2-4. Derivations 9.2 The automorphisms of a Lie algebra G form a linear group, denoted by Aut G. Though the Lie character of Aut G still has to be established (see 11.3.4), the Lie group notions can be applied to Aut Gin a heuristic approach. Definition

IR is called an injinitesimalautomorphismof Gif

for some curve )',0, in Aut G with 0, = 1. With the notation of this definition one gets @,[a,b]= [@,a,0,6];hence by differentiation at t = 0 52[a,b] = [IRa,b]i [a, Qb]. Clearly52 is linear. This suggests the following. 9.3. Definition A linear mapping IR of G into itself is called a derivation of G if Sl[a,b]= [Qa,b] [a, Qb].

+

It has been shown that an infinitesimal automorphism is a derivation. The converse is also true:

9. HOMOMORPHISMS, AUTOMORPHISMS, DERIVATIONS

37

Proposition The notions of infinitesimal automorphism and derivation of a Lie algebra G coincide. If Q is a derivation, exp s;! is an automorphism. Proof

Let s;! be a derivation of G. Put c, = [(exp fQ) a, (exp fQ) 61. Then d

- c, =

dt

[Q(exp tQ)a, (exp rQ)b] + [(exp tQ)a, Q(exp tQ)b]

= Q[(exp

&)a,

(exp tQ)b]

=QCt.

The differential equation

admits the unique solution c, = (exp tQ) [a,b].

This proves exp tQ

E

Aut G.Because of ($exp

tn)

=Q,

t=o

Q is an infinitesimal automorphism. 9.4. Proposition The derivations of a Lie algebra G form a linear Lie algebra. Proof Clearly they form a linear space. LetQ,,Q2 be two derivations of

G.

. n , Q , [ a , ~ l = [ Q , Q , a , b ] +[ Q , a , Q,bI+[Q,a, Q , b I + [ a , Q,Q,bI. [Q,, Q21[a,61 = QlQ2 [a,61 - Q2Q, [a, bl = [(Q,Q2 = “Q,,

-Q,Q,)a,bl

Q 2 1 a, 61

+ [a, 1). The subgroup of Aut Rz,+,characterizedby det = 1 and the invariance of a nondegenerate quadratic form s = s’ (in matrix notation). The Lie characterized by u’s + sa = 0. subalgebra of a E End C, (1 > 1). The subgroup of Aut Rz, characterizedby the invariance of a nondegenerate skew bilinear form, in matrix notation s = -s’. The Lie subalgebra of a E End Rzl, characterized by a’s sa = 0.

+

D, (1 2 2). As B,, with 21 instead of 21

+1

I

These groups belonging to A,, B,, C,, and D, are also called the special (1 + 1)-linear group, special (21+ 1)-orthogonal group, 21-symplectic group, special 21-orthogonal group. The dimensions are A,: (I+ 1)’- 1; B, and C, : 1(21+ 1); D,: 421- 1).

The representatives with which these classes of groups have been introduced are linear Lie groups. For a few of them it has been proved in 11.3.4 and for the others the proof is analogous. 16.2 G E A,. Let H be the (maximal abelian) subalgebra consisting of the diagonal matrices with trace 0. The diagonal coefficients in due order are functions defined on H,denoted by ul, ...,w,+,, with I;u,(h) = 0 for h E H

16.

IMPORTANT CLASSES OF COMPLEX LIE ALGEBRAS

77

or, for short, w , =O. The elements e,, (i # j ) are defined as bearing 1 on the i-jth place and 0 elsewhere. Together with a basis of H, they form one of G. The structure of G is described by the commutator relations:

H abelian, [h,el,] = ( w , - w , ( N e,,, for [e,,,e,,l = 0 for i # q , P # j , re,,, eJkl= elk for i # k , [e,,, ekll = -ekj for j # k, [el,, e,,l = hi, E H

E H,

where wp(h,,)= 1 for p = i, = -1 for p =j , = 0 for p # i, j . H is no ideal in a larger subalgebra of G. Anticipating a more general terminology, H i s called a trunk of G.

h,= ( w , - w,) kho=0

e,,, for ho E H ,

for h E H, show that the roots of h are ( w i - w,)(h) and I-fold 0, with simultaneous eigenvectors of h: e,, and nonvanishing elements of H. The restrictions of the roots to Hare called the rootforms (with respect to H). They are linear functions on H, and are simple, except 0, which is I-fold. Among them there are 1 linearly independent ones : w, -0 2 , 0 2

- w3,

. .., w , - w,+,.

Every rootform is a linear combination of this subset with integral coefficients, all > 0 or all G 0.

I = number of linearly independent rootforms = number

of functionally independent rootforms G number of functionally independent roots = rank of G (see 15.3) Q nullity of G (see 15.5) G multiplicity of rootform 0 = 1.

This shows that rank G = 1. The negative of every rootform is again a rootform. The eigenvector e,, (determined up to a scalar factor) of the rootform w , - w, is called the brunch (with respect to H) belonging to the rootform

78

13-19.

SOLVABILITY A N D SEMISIMPLICITY

w , - u,.The commutator of branches belonging to opposite rootforms is called the node belonging to them (determined up to a scalar factor); /I,, is the node belonging to w 1- w,. ForhEH tr RZ = 2 ( w , - w , ) (/I). ~ tr R = 0,

z

I 0 or all s 0. The negative of every rootform is a rootform. The rank is again 1. The commutator relations are somewhat involved. The following general principle is useful : If e, e‘ belong to the rootforms a, a’, then [e,e’]either belongs to the rootform a + a’ or vanishes. Indeed, from [h,e‘] = d ( h )e’, [h,el = a(h)e, one obtains by Jacobi-associativity

[h, [e, e l l = “h, el, ell+ [e, [h,e’ll = ( a

+ a‘)(h)[e,e‘l.

From this it follows that [e,e’] E H as soon as a + a’ = 0 because H is maximal abelian. Furthermore, for any e,e’ chosen from the set of branches elj, tr IC‘ = 0, [e,e’] = scalar multiple of e”,

+ a’ # 0 is a rootform and e” its branch. Let 1 s i, j G 1, i # j . The pairs ell,ej, and el,j+l,ej+l,lbelong to opposite rootforms and their nodes are hlj E H with 1 on the ith and ( j + 1)th places, -1 on the jth and (i+ I)th places, and hi,j+lE H with 1 on the ith and jth places, and -1 on the (i l)th and ( j + 1)th places. On H the Killing form tr R2 = 4(1- 1) wj(h)2 if

CL

+

is nondegenerate and so is the total a,h, which can be shown by a somewhat longer computation.

80

13-19.

SOLVABILITY AND SEMISIMPLICITY

16.4 G E B,. The quadratic invariant is assumed to be

to2 + X I &+,

+

* *

+ 2&t 2 b

with the matrix

a horizontal and a vertical line being added to the s of 16.3. The elements of G have the form -a; a2

-a; -a1

a3

with a2 + a; = a3 + a; = 0. A trunk His again the set of diagonal matrices with the values of 0,w1, . ..) w,, -w1,

* *

., -w,

as diagonal coefficients. To the branches of D, are added eoi (1 Q i Q I ) with 1 on 0, i; -1 on i + 1, 0, qo (1 Q i Q I) with 1 on i, 0; -1 on 0, i + 1. They belong to the rootforms -w,,w, so that kw,, fw,f wJ (i # j ,

independent) and 1-fold 0

are all rootforms. A basis with the old properties is provided by w1 - w2, w2 - w3,

. ..,

W,-l

-w,, w,.

The rank is 1 and all the other properties reappear. 16.5 G E C,. The skew bilinear invariant has the matrix s = ( -1O

01).

where 0, 1, -1 represent I-l-matrices. From

and a's

one gets

+ sa = 0

17. a* = a;,

81

SOLVABLE SUBALGEBRAS

a,

= a;,

+ a;

111

= 0;

a2 and a, are symmetric instead of skew, as they were i n D,. After the usual choice of a trunk H , this means branches and rootforms like those of D,, though in the second and third kind the matrix coefficient -1 should be changed to 1 ; furthermore the branches ei,i+,

with 1 on i , i + I,

with 1 on i + I,i,

and zeros elsewhere, with the rootforms 2wi, -2wi, must be added. All rootforms are *2wi, &wi

*

wj

(i # j , f independent) and I-fold 0.

A basis with the usual properties is provided by w,

-

w * , w* - w3,

. . .) w l - ,

- wi, 2w,.

The usual properties can be verified again. 16.6 The properties displayed in theseexamples arecommon to the important class of complex semisimple Lie algebras. Their study is pursued in Sections 20-27. 17. SOLVABLE SUBALGEBRAS

G E Alg Lie (generally Com, sometimes Rea), dim C = I' < cc ; R Com, dim R = IT < 3 3 .

E

Spa Lin

17.1-3. Nilpotency 17.1. Definition An element u of End R is called nilpoteiit if up= 0 for some positive integral p. A subset A of End R is called iiilpoterit if A P = 0 for some natural p. (Here A p is defined inductively by A' = A , A4+I = AAq.) Evidently an element is nilpotent iff all its eigenvalues are zero. If Gi is defined as the set of complex 17-n-matrices with zeros under and i n the main diagonal and in the first i - 1 parallels above the main diagonal (i = 0, I , . . .), then Gi cj = Gi+j (see 13.11, Proof). Thus GI is nilpotent. Definition a E C ( A c C ) is called ad-nilpotent if B ( A ) is nilpotent. An element is ad-nilpotent iff all its roots vanish. All elements of C a r e ad-nilpotent iff G is of rank 0.

82

13-1 9.

SOLVABILITY A N D SEMISIMPLICITY

Proposition Let G c End R be a complex linear Lie algebra. If every element of G is nilpotent, then G is solvable and even nilpotent.

Proof The statement is supposed by induction to be true for any R’instead of R and any G’ (cEnd R’) instead of G as soon as either dim G’ < dim G or dim R‘ < dim R . Let H be a maximal proper subalgebra of G. Then every element of H is nilpotent, so by induction H is solvable. According to Lie’s theorem (13.9) take x E R as a simultaneous eigenvector of H. Because of the nilpotency of

the elements of G

17.1.I

HX= (0).

Let K be the maximal subalgebra of G such that Kx = {O}. Then K = G or If Gx = {0},then G acts upon R‘ = R mod x as an algebra G‘ which by induction is nilpotent; this implies the nilpotency of G. If K = H then ax # 0 for all a E G\H. H is invariant under h with h E H. Thus h induces a linear mapping B(h) of Gmod H into itself; B is a homomorphism; B(H), as a homomorphic image of H,is solvable. According to Lie’s theorem there is a simultaneous eigenvector of i!(H).Thus for some u E G, u f H , = H.

17.1.2

-

ha = a(h)a mod H

Now has

=

[h,a] s

=

a(h)US

for all 1 E H .

+ ahs

because of 17.1.1-2. The nilpotency of h requires a(h) = 0, which by 17.1.2 shows that the algebra spanned by H, a, hence G, is solvable. I t may be assumed to be triangular, but then it has zeros in the main diagonal and so it is nilpotent. Corollary A complex linear Lie algebra has rank 0 if and only if its adjoint

is t i i I po ten t .

17.2 A solvable linear Lie algebra need not be nilpotent. Its commutator algebra, however, is nilpotent because the diagonal of the commutator of triangular matrices vanishes. Proposition A complex linear Lie algebra is solvable iff its commutator algebra is nilpotent.

17.3 If G is a Lie algebra, the preceding propositions can be applied to the linear Lie algebra G.

17.

83

SOLVABLE SUBALGEBRAS

First Criterion on Solvability The complex Lie algebra G is solvable iff its commutator algebra C ( G )is of rank 0 (ad-nilpotent). Proof If G is solvable, then by 17.2, for any u E C ( G ) , all eigenvalues of &

vanish and, in particular, all roots of u as element of C ( G ) . Conversely, if rank C ( G ) = 0, its adjoint is nilpotent, and therefore solvable, and so are C ( G )and G. (See 13.2.3.) 17.4-1 2. Trunks 17.4 Let G E Alg Lie Com, ho E G. According to the different roots of h,, G as a linear space splits directly into linear subspaces G, such that ( K O - a)d'mCa G, = (0).

Proposition [G,, G,] c G,,, or = (0) (if a Proof For x

E

+ /3 is no root of ho).

G,, y E G,, and some i,j

(Lo - a)' x = (Lo - p)'y

= 0.

Now (60

-a -

B) [ w l = [(KO

- a)x,ul+ [x, ( 6 0 - B)Yl.

Iterating this, one gets

(Lo - a - /3)'+'-'

[x,yI= a sum of [(KO -

(Lo - ,QqyI,

with p + q = i +j - 1, hence p > i or q j . This means that all summands vanish and proves the assertion. 17.5 In particular, [Go,Go]c Go,which shows that Go is a subalgebra. 17.6. Definition

For any regular ho E G the set of all x E G such that &X=O

forsomep

is called the trunk of Lo. The letter H usually denotes a trunk. Theorem A trunk H of G E Alg Lie Com is a solvable subalgebra of rank 0; H i s the trunk cf all its G-regular elements and contains all simultaneous eigenvectors of 6 (h E H) for the eigenvalue 0. It is an ideal in no larger subalgebra of G. Proof H, the trunk of regular ho, called Go in 17.5, is an algebra and

therefore invariant under

with h E H. The multiplicity of the eigenvalue 0

13-19.

84

SOLVABILITY AND SEMISIMPLICITY

ofh, as acting onHand GmodH, is denoted by nH(b)and nmdH(b),respectively. Continuity prevents &,,OdH(b) from increasing near b,; regularity prevents the nullity of 6, that is, nH(b)+ n m o d H ( b ) , from decreasing. Therefore nH(b)2 nH(b,)for b near b,. Hence

RPx= 0

for x E H and some p

and for all /I E Hnear b,, but because of the analyticity of hPx as a function of b this holds good for all b E H. All b E H are ad-nilpotent as elements of the subalgebra H;by 17.1 the same is true of H, which, again by 17.1, is also solvable. The trunk of any regular /I E H i s at least H. It cannot be larger because this would enlarge the nullity of b. The third assertion of the theorem is a mere consequence of the definition of trunk. If H were a proper ideal of some subalgebra of G, there would be some a $ H such that ha E H for all b E H and hp"a = 0 for some p, which would produce the contradiction a E H. 17.7 A converse of 17.6 is proved in 17.1 1: Theorem Let A be a subalgebra of rank 0 of the complex Lie algebra G and let A be ideal of no larger subalgebra of G. Then A is a trunk of G.

17.8t Under automorphisms of G a trunk goes into trunks. In 17.1 1 it will be proved : Conjugacy Theorem Two trunks of complex G are equivalent under the

action of Int G. 17.9 The proofs are prepared by means of the next propositions. Proposition Let G E Alg Lie Rea or Com, A a linear subspace of G, a, E A such that d , A c A and the mapping induced by do in the linear space G mod A is nondegenerate. Then the (exp b)A with b from a O-neighborhood in G cover a neighborhood of a,. If, moreover, G E Alg Lie Corn, then A contains a regular element of G. Proof As a linear space, G is the direct sum of A and a linear subspace B. The

mapping 6 defined by 9 ra,bl

= (expb)(ao

+ a)

for a E A, b E B

t 17.8-12 will not essentially be used for some time.

17.

85

SOLVABLE SUBALGEBRAS

has at ‘0,Ol (grad 6) ru, b 1 = u

+ [b,a,]

=u-

b.

Now u - d o b = 0 would mean a,b = 0 mod A , which by the supposition on nondegeneracy implies b = 0, hence u = 0. Therefore grad 6 at ‘0,O’ is nondegenerate and 6 maps a neighborhood of ‘0,O’ onto a neighborhood of uo. This proves the first assertion. The set of regular elements is dense, so the neighborhood of uo covered by the (exp6)A ( b e l l ) contains a regular element. Since (expd)A possesses a regular element for some b, and the notion of regular element is Int G-invariant (see 15.7), A has the same property. 17.10. Proposition Let G E Alg Lie Com, A sub G, rank A = 0. Let the mapping induced by d (a E A ) in the linear space C mod A have an identically vanishing eigenvalue. Then A is an ideal in a larger subalgebra of G.

Let h, E A be chosen with a minimal nullity in G among elements of A . As in 17.4, G, is defined with respect to h,. Then, since rank A = 0, A c Go. The postulated behavior of h, in G mod A implies the existence of x E C A \ such that h,x E A , hence x E Goand Go # A . By 17.4 Go is invariant under A,. The restriction of h, to Go is nilpotent, and the same reasoning used in 17.6 shows that this property remains true for all h E A because of the minimal nullity ofh,. So A is represented nilpotently in GobyyhpAh,and so it is in Gomod A . Let x + A E Go mod A be an eigenvector of this representation (with the eigenvalue 0). Then the linear subspace spanned by A and x is a subalgebra properly containing A as an ideal. Proof

17.11 Proofof Theorem 17.7-8. Let A be assumed as in 17.7. Then by 17.10 no identically vanishing eigenvalue of the mapping induced by a (a E A ) in

G mod A exists. By 17.9 A possesses a regular element h, of G, the trunk of which is called H . Since rank A = 0, the restriction of h, to A is nilpotent, and since H itself is the maximal linear subspace with this property, A is contained in H . If A # H , then A, acting nilpotently on H m o d A , has an eigenvector A\ such that Ax = belonging to the eigenvalue 0. Hence there is some x E H (0). The subalgebra spanned by A and x contains A as an ideal. This contradiction shows that A = H, hence that A is a trunk. For any trunk H let @ ( H ) be the set of regular elements of (Int G ) H . By 17.9 (which applies because of 17.10) @ ( H ) is open, and by the same reason If there were several @ ( H ) , the every regular element belongs to some @(H). and set of regular elements, which is connected (15.4), would split into @(H,) a union of certain @ ( H ) ,which are both open. This contradiction proves 17.8.

86

13-19.

SOLVABILITY AND SEMISIMPLICITY

17.12. Theorem Let cp be a homomorphism of G E Alg Lie Com onto G' E Alg Lie Com. By cp the regular elements of C are mapped into regular elements of G', and the trunks o f G' are the images of the trunks of G. Proof Let N be the kernel of cp; G' is canonically isomorphic with G mod N.

If h, is a regular element of G, then cp(h,) is one of G' because the regular elements of C' are dens,ein G'and a slight variation of ho does not change its nullity on N, thus cannot decrease its nullity on G mod N. It is clear then that the trunk H associated with a regular element ho of G is mapped onto the trunk H' associated with cp(ho).Any other trunk of G' is of the form#'H' for some $' E Int G'. Now Int G' is canonically the image of Int G, and, if t,h E Int G is mapped into #', then #H is a trunk of G mapped onto I)' H' by cp. 17.13-15. Rootforms and Nodes 17.13.1 Definition Let H be a trunk of G E Alg Lie Com. The restrictions of the roots to H are called the rootforms (with respect to H). The system of rootforms, with due multiplicities, is denoted by W or, more precisely, W(G,HI. Since H is solvable, its respresentation by y h h may be considered triangular, with the rootforms in the diagonal. Thus: Theorem The rootforms are linear functions on the trunk.

17.13.2 The set of h E H such that a(h) = p(h) for some couple a, @ ofdifferent rootforms is nowhere dense in H. One can choose some h, E H such that for any couple a,fl of rootforms a(ho)# fl(ho)unless a = fl. Splitting G into G, (see 17.4) with respect to h,, one obtains

(6,

- .(ho))d'"C"Ga = (0).

Because of 17.4, G, is invariant under h (h E H). The restriction of h, to G, has one eigenvalue only. With a view to the choice of A,, it appears that this property belongs to all h E H . Hence : Theorem Let H be a trunk of G E Alg Lie Com. Then G splits as a direct

sum of subspaces G, corresponding to the different members a of W such that

(h - a(h))d'mGaG ,= (0) for h E H ; [G,,Gp]c G,+p or (0) (if a + fl is no rootform). Gois the same as H. 17.14 By 17.8, W ( G , H )is essentially the same for all trunks H of G. Any functional dependency among roots causes a dependency among the corre-

17.

87

SOLVABLE SUBALGEBRAS

sponding rootforms, and conversely any dependency among rootforms is raised to a dependency among roots by the action of Int G. Functional dependency among rootforms coincides with linear dependency. Therefore: Theorem The rank of C equals the number of linearly independent rootforms. The last-mentioned number does not exceed dim H, which is the nullity of C . This confirms Theorem 15.5. 17.15. Definition If a and -a belong to W, the nonvanishing elements of [G,,C-,] are called the a-nodes (with respect to H ) . Of course, the cr-nodes belong to H . Theorem For any a-node h and any rootform p,p(h) is a rational multiple of a@), where the factor does not depend on h (though it may depend on a and p). Proof Let h = [f+,f-] where f+

E

G,, f- E G-,. Then

as far as existent. Put

F=j

C

Gp+jo,;

integer

F is invariant underf+,r-, thus under 6. Since 6 is a commutator of elements preserving F, trF6 = 0; however, trF6 = Z I trG,,+,,6. In Gp+Iathe only eigenvalue of 6 is (p +ja)(A) with multiplicity dim Gp+I,.Thus,

C dim Gp+Ja.(p+ j a ) ( h ) = 0. I

This proves the assertion. 17.16-17. The Killing Form and Solvability 17.16 If E sub G one can restrict the Killing form to E and then distinguish between $ E ( x , Y ) = t r E fj (x, Y E )

and $G(x, Y ) = trG f$

(x, Y E E),

which in general will be different. However: Proposition If E is an ideal in G E Alg Lie, then 4E(x,Y ) = #G(x, Y )

for x, Y

E-

13-19.

88

SOLVABILITY AND SEMISIMPLICITY

Indeed, if E is an ideal, then 2 j z E E for x,y butions to the trace come from z E E only.

E E, even

if z $ E. So contri-

17.17 If G E Alg Lie Com is solvable, then, according to the first criterion on solvability (17.3), all roots of elements of C(C) vanish [not only on C(C) but also on G because C(G)is an ideal]. As the sum of the squares of the eigenvalues of 2 , $(x,x) then vanishes for all x E C(C).The converse is also valid: Second Criterion on Solvabi!ity G E Alg Lie Corn is solvable if and only if i,b(x,x) = 0 for all x E C(G). Proof One still has to show that if $(x,x)

=0

for all x E C(G) then G is

solvable. By 17.16 the supposition implies trEf2=0

for x E E,

where E = C ( E )is the last member of the commutator sequence of G. It has to be shown that E = (0). Suppose that E # (0) and choose a trunk H of E. Then H # (0). As a trunk H is solvable, whereas E = C ( E ) is not, and so there exists a rootform # 0 with respect to H.Since E= C ( E ) ,the trunk H i s spanned by the nodes; therefore there is a node ha of some a such that P(ha) # 0. Then, according to 17.15, 0 = $(h,, h,)

dim C,,~ ( h , =) Ka(ha)’ ~

= P

with K # 0. Therefore a(ha)= 0 and again, by 17.15, p(h,) = 0 for any rootform p, in particular for p, which is a contradiction. 17.18. The Real Aspect By 13.13 and 14.6 the notions of solvability,

commutator algebra, and vanishing Killing form are invariant under complex extension, real restriction, and waiving. These properties are also shared by the notion of being of rank 0. The two solvability criteria remain valid in G E Alg Lie Rea. 17.19. Historical Note Theorem 17.12 is due to H. de Vries.

18. CLEAVING 18.1 R E S p a L i n C o m , d i m R = n c m . Definition A subset and in particular, an element of End R is called pure if it takes the diagonal form on a suitable basis.

18.

CLEAVING

89

Definition IfhEEnd R, the splitting h = a + e is called a cIeauingifaEEnd R is pure, e E End R is nilpotent and ae = ea. 18.1.I. Proposition h E End R admits one and only one cleaving. The cleaving components are scalar linear combinations of the powers of h with positive integral exponents. As a consequence : Every u E End R commuting with h also commutes with the cleaving components. Every linear subspace invariant under h is invariant under the cleaving components. Even: if S , T sub R and hS c T c S , then aS c T, eS c T. Proof R splits directly into subspaces R, corresponding to the different eigenvalues A,, . . ., A, of h. In every R, choose a basis on which the restriction ofh to RI is triangular. Take the union of these bases as one for R. The diagonal part of h is taken as a, the remainder as e. Obviously this is a cleaving. Conversely, if the cleaving h = a e is given and Xi are the different eigenvalues of a with eigenspaces Ri,

+

(a - A,)

R*= w,

then

(h - A,)" R, = ((a - A,) + e)"R,, which by the binomial formula and by en = 0 turns out to be (0). This shows that the X i are the eigenvalues of h and the R, the corresponding subspaces, and that the cleaving coincides with the previous one. Further define a polynomial y ( z ) such that for ~ ( zand ) its derivatives y'(Ai) = * . . = y+')(A,) = O for i = 1, . . ., p . y(AJ = A,, y(0) = 0, The polynomial y ( z ) - A, then has a factor ( z - A,)". Substituting h = a e into v, one gets y(h) = a. This is a linear combination of hi (i > 1) as required.

+

18.1.2t Cleaving as defined in End R has a multiplicative analog in Aut R.

Proposition h E Aut R can be split uniquely as h = au where a is pure, u is unipotent (i.e., all its eigenvalues are l), and au = ua. This follows from the cleaving h = a + e with pure a, nilpotent e, and ae = ea by putting u = 1 + a-'e. 18.1.3 Let D be the set of h E End R such that all eigenvalues X of h satisfy

< Im A < rr, and D, the subset of pure elements of D . Then exp is one-to-one on D,. -IT

t

18.1.2-5 will not be used before Section 38.

90

13-1 9. SOLVABILITY AND SEMISIMPLICITY

exp maps the (additive) cleaving of h E End R on the multiplicative one of exp h ; moreover, it maps the set of nilpotents onto the set of unipotents, since exp is inverted by log. Therefore exp maps D one-to-one. 18.1.4 Now interpret End R as a linear Lie algebra G. The kernel of the trace in End R is a member Goof An-,. For a E G let a, be such that a - uo is a scalar multiplier. The eigenvalues of a' or, equivalently, of do are the roots of ao, hence are differences of the eigenvalues of a, or, equivalently, of a if a, belongs to a trunk of Go.Hence by a continuity argument this, however, holds in general. So for a E D and all roots a

-2r < Im .(a) < 27r; hence, by 10.2, grad, exp is nondegenerate. Together with 18.1.3 this yields the following. Theorem exp maps D homeomorphically and with nowhere degenerate gradient onto an open subset of Aut R, which is an analytic manifold.

18.1.5 An application. Proposition If G is the linear Lie algebra of the real n-n-triangular matrices of the form then exp G coincides with the group infinitesimally generated by G (which, in fact, is the group of real n-ti-triangular matrices with positive eigenvalues), and exp is a homeomorphism of G onto exp G. The same holds for subalgebras of G as well.

A,

18.2 The foregoing is applied to the adjoint of G E Alg Lie Com. Consider the splitting of G into G, under some h E G, as introduced in 17.4. Then

(h - a)d'mCaG,= {0},

[G,, Gp] c G,+p

or (0).

Define a linear mapping 9. of C into itself by 8x = a x for x E G,. Then 6 is a derivation of G, for if x EG,, y E G ~ , then [ x , y ] E G , + ~ or 0; hence 9.[n,yl = ( a

+ 8)[ X , Y l =

[ax,yl+ [x, PYl

= [9.X,Yl+

[x, 9.Yl

or 0 in the first and last members. Since 9. is the first component of the cleaving of k,one is led to the following.

18.

91

CLEAVING

Proposition If G E Alg Lie Coni, the cleaving components of h (h E G ) are derivations of G. 18.3 A stronger assertion can be made ifG is centerfree and all its derivations are inner. (Such is the case for the seniisitnple algebras to be studied later on.) Definition h E G E Alg Lie Corn is called ad-pire if of G is called ad-piwe if the set ofu with a E A is pure.

h

is pure. A subset A

+

Definition For h E G E Alg Lie Coni, I1 = a e is called an ad-clearing if a,e E G, a is ad-pure, e is ad-nilpotent, and [ q e ]= 0. Theorem Let G t Alg Lie Coni such that C is centerfree and has inner derivations only. Then every element k of G admits unique ad-cleaving. If h belongs to the trunk H , then the same is true of its ad-cleaving components.

According to 18.2, the cleaving components of h are derivations and thus inner derivations of the form cS.,P with [ B , P ] = 8. Then [aTe] = 8 , hence [a,e]= 0 because Giscenterfree. This proves the first part. If li E H , h leaves H invariant; by 18. I . I the same is true of its ad-cleaving components; by the last remark of Theorem 17.6 their ad-originals belong to H .

Proof

18.4. Proposition If cleaving is possible in G and tlie trunk H does not possess ad-nilpotents # 0, all elements of H are ad-pure. A direct consequence of the proof of 18.3. 18.5t Suppose that R E Spa Lin Com, dim R = I I < . a,A c End R,A pure. Then the linear span of A is also pure. Call R , the subspace belonging to the eigenvalue o! (as a function on A ) . Suppose that S sub R is invariant under A . Then tlie following is easily shown by the method of 18.1. 18.5.1 S is the direct sum of the S n R,. 18.5.2 The restriction of A to S is still pure. 18.6. Proposition If R E Spa Lin Com, A u B c End R, A u B abelian and both A and Bare pure, then A u B is pure. Proof

R, is again defined with respect to A . Since A u B is abelian, R, is

B-invariant. The restriction of B to R, is still pure. For every o! take a basis of t 18.5 will not be used for some time.

92

13-19. SOLVABILITY AND SEMISIMPLICITY

R, on which B is diagonal and combine them. This provides one on which A u B is diagonal. 18.7. Historical Note The importance of cleaving in semisimple Lie algebras was discovered by F. Gantmakher, Mat. Sbornik 5, (47), 101-144 (1939).

19. SEMISIMPLICITY

First read the criteria and theorems. G E Alg Lie, dim G = r < 03. 19.1-1 3. Criteria 19.1 Repetition of parts of Section 13. Definitions The maximal solvable ideal of G is called the radical, rad G, of G. G is called semisimple if rad G = {0), equivalently, if G has no abelian ideal except (0). Notation If G is semisimple: G E Alg Lie SS. If G is semisimple and simple: G E Alg Lie SSS. If G E Gru Lie Lin the predicates SS and SSS indicate that G E Alg Lie SS and SSS, respectively. It has been proved that if G is simple and dim G > 1,then G E Alg Lie SSS. G mod rad G E Alg Lie SS.

19.2 The Killing form of G is denoted by $. This is a &invariant (see 14.4.1).

-

Notation For K c G the $-orthoplement of K is written K L ;thus, u E KJ-

$(u,K) = (0).

19.5 From 19.5.1

$(ex, Y ) + $(x, ZY) = 0

it follows : Proposition IfFis an ideal in G, the orthoplement FJ-of Fwith respect to Q, defined by Y E FJ-tf A ~ ( x , Y= ) 0, XEF

is again an ideal of G.

19.

SEMISIMPLICITY

93

19.6 In particular, the set G1 of ally such that $(x,y) = O

for all x E G

is an ideal. By 17.16 +(x,x) = $,i(x,x) for x E G I , and by the second criterion on solvability G1 is a solvable ideal of G. Therefore, if G is semisimple, the only solution of for all x E G

$(x,y) = O

is y = 0, which means that $ is nondegenerate. If, however, G is not semisimple and F # (0) is an abelian ideal of G, then f j z = [x, [y, z]]

EF

for x, z E G, y E F.

If z contributes to tr fj, then z E F, but then [y,z] = 0; thus f j z = 0. Consequently, tr 29 = 0, $(x,y) =0 for x E G, y E F, which means that $ degenerates as soon as G is not semisimple. This yields the: First Criterion on Semisimplicity G is semisimple iff its Killing formis

nondegenerate. 19.7 Let F be an ideal of G and let F1 be defined as in 19.5. If dim G = r and x,, . . ., x, form a basis of F,then FL is the intersection of the linear subspaces

-

L j of at least (r - 1) dimensions, defined by y This makes dim Fl

E

L,

$ ( X j , Y ) = 0.

r - s, dim F + dim F1 > dim G ;

F n FL is an ideal of G and is solvable because of 17.16 and the vanishing of t,h on F n P. 19.8 Applying this result to a semisimple ideal F of G, one gets the ideal Fl (see 19.5) with F n FL = (0) and F + FL = G. Theorem If a semisimple Lie algebra is an ideal in a larger one, it is so in a

trivial way, that is, as a direct summand. If a semisimple linear Lie group is a normal subgroup in a larger one, it is so in a locally trivial way, that is, as a local direct factor. 19.9 A consequence of 19.8:

94

13-19.

SOLVABILITY AND SEMISIMPLICITY

Theorem All derivations of semisimple F a r e inner. Proof

Let G be the linear Lie algebra of derivations of semisimple F and let

E be that of the inner derivations; P is an ideal in G since, for @ E G ; [@, ci]x = @ax - a@,x = @ [ a , x ]- [a,@x]= [@a,x] + [a, @XI - [a, @XI = @ a x ; hence: N

v

19.9.1

[@,&I = @a,thus [@,PI c F.

Since Fis semisimple, its center is (0) and Pis isomorphic to F ;consequently, it follows it is semisimple. By 19.8 G is the direct sum of P a n d Pl. For @ E N

from 19.9.1 that 6 G = P.

2

[@,a]= @a;thus @a = 0 for all a E F. Hence @ = 6 and

Corollary For any x G. (See 18.3.)

E

G E Alg Lie Com SS cleaving of 2 is possible within

19.10. Theorem The 1-component of Aut F of semisimple F is Int F. Both are closed in the group of linear space automorphisms of F. Indeed, by 11.3.4, the infinitesimal algebra of the 1-component of Aut F consists of the derivations of F, which by 19.9 are all inner and infinitesimally generate Int F. The closedness of Aut F is obvious and implies that of Int F.

In 38.5 it will be shown that any G Spa Lin is closed in Aut R.

Remark

R

E

E

Gru Lie Lin SS acting in

19.11 A consequence by means of 9.1 is the following. Theorem The automorphisms of G E Gru Lie Lin SS i n the l-component (even the local ones) are inner; Aut G/Int G is discrete.

19.12 Now apply 19.7 to the case of a semisimple G and an ideal F. Again F n FL as a solvable ideal of G is {O} and G is a direct sum of F and FL,which are

again semisimple. Repeating this process, one obtains the following: Second Criterion on Semisimplicity G is semisimple iff it is a direct sum of nonabelian simple Lie algebras. The sufficiency is obvious. Note that the splitting of semisimple G is uniquely determined. Indeed, any simple ideal other than the summands would have zero-intersection with each of them and therefore would commute with each of them elementwise; thus it would be contained in the center, which is {O}. Globally :

19.

SEMISIMPLICITY

95

E Gru Lie Lin is semisimple iff it is locally isomorphic to a direct product of locally simple nonabelian linear Lie groups.

Theorem G

19.13 From 19.12 it is evident that any ideal and any homomorphic image of semisimple C is semisimple. Proposition

If G’ is an ideal sub C, then rad C ’ is still an ideal sub G.

Proof This follows most easily by observing that any rp E Int G induces an automorphism of G ’ , which consequently maps rad G’ onto itself. In a purely algebraic fashion from the semisimplicity of G mod rad G one draws the conclusion that C ‘ mod G’ n rad G, isomorphic to an ideal of the former, is also semisimple; thus rad G’ c G’ n rad G, which is a solvable ideal of G’, hence equals rad G ‘ .

19.14-1 6. Irreducible Algebras 19.14 Because of 19.12 semisimple C i s clearly its own commutator algebra. Thus, on G E Alg Lie Lin SS the trace function vanishes.

Let R 6 Spa Lin Com, dim R < co. If G E Alg Lie Lin acts irreducibly on R , it splits directly as a sum of a semisimple ideal Go and a Lie algebra of scalar multiplications. This re-enforces 13.10.

Theorem

Let Go be the set of g E G with tr g = 0. Then Gois an ideal of G.By 19.13 the same is true of rad Go, which according to 13.10 consists of scalar multiplications. Since tr(Go) = {0},it follows that rad Go= (0) and that Gois semisimple. According to 19.8, G E Go + A as a direct sum. The commutator algebra of C is in the kernel of the trace, thus in Go,from which it follows that A is abelian, and thus by 13.10 consists of scalar multiplications. Proof

19.15 The following is an extension of 19.14:

Let R E Spa Lin Com, dim R < co, be the direct sum of R, sub R . If G E Alg Lie Lin acts on R, leaves the R, invariant, and acts irreducibly on each R,, then G is a direct sum of simple Lie algebras. Theorem

Proof The restriction of G to R, is a linear Lie algebra Gj, homomorphic

image of G by means of canonical 6,. The kernels F, of 5, have intersection (0). Let C ’ and C; be the commutator algebras ofG,G,. Then 5,C’ = G;. By 19.14 G; is semisimple, 5, rad G’ c rad G; = (0). Thus rad G’ = {0}, G‘ is semisimple,

96

13-19.

G = G’

SOLVABILITY AND SEMISIMPLICITY

+ A as a direct sum and, because of 19.8, A abelian. This proves the

assertion. 19.16. Theorem All G E A,, B,,C,, D, of Section 16 are semisimple. Because of 13.8 it has only to be shown that the representatives G are irreducible. Since an invariant linear subspace would also be invariant under the associativealgebra with unit element generated by G, it suffices to show that the latter is irreducible, that is, the full matrix algebra of complex d-space on which G acts. The trunk of G, together with the unit matrix, generates all diagonalmatrices, and by multiplyingthe diagonal matrices ‘0, . .,0 , l ,0, ...O1 by branches of G one gets all branches of Ad. In this way the full matrix algebra of complex d-space is produced.

.

19.17. The Real Aspect Because of the validity of the solvability criteria for real Lie algebras the results of Section 19 prevail for real as well as complex Lie algebras. This is especially true of the semisimplicity criteria. Semisimplicity is preserved under complex extension, real restriction, and waiving. With a slight modification 19.1415 can be extended to the case of real R. Then, of course, G has to be real. To tackle 19.14 one passes to the complex extension Rcom of R, which as a matter of fact can become reducible. Together with S # (0}, R, its conjugate DS is invariant sub Rco,. The real restrictions of S n DS and S DS are invariant sub R, and thus = (0 ) and R, respectively, which shows that Rcom= S DS.To this situation, however, the method of 19.15 readily applies. In the formulation of 19.14 for real R the scalar multiplications must still be allowed over the complex field.

+

+

19.18-22. Purity and Cleaving 19.18 If G E Alg Lie is centerfree, then G is isomorphic to G.Then, up to isomorphy, Gmay be assumed to be linear. This remark applies in particular to semisimple G. 19.19t. Theorem Let G E Alg Lie Corn Lin. Then every pure element of

C is ad-pure and every nilpotent element of G is ad-nilpotent; in particular, if G admits cleaving, G also admits ad-cleaving; therefore if G is centerfree and admits cleaving, then for elements of G the notions of purity, nilpotence, and cleaving coincide with those of ad-purity, ad-nilpotence,and ad-cleaving.

t

19.19-22 will not necessarily be used for some time.

19.

97

SEMISIMPLICITY

Proof G c End V with V E Spa Lin Com.

(End V ) , is the kernel of the trace of End V. One may suppose that dim V > 1. First assume that G = (End V),. Then by 19.16, G is semisimple. Since the cleaving components of a linear mapping with trace 0 also have trace 0, G admits cleaving. From 16.2 it is easily seen that G contains a pure trunk; in fact, every pure element of G is in a trunk of G. The elements of the given trunk are ad-pure; hence the elements of any trunk are ad-pure because the trunks are conjugate under G (if G is the group of linear mappings with determinant 1) and G" preserves ad-purity. By putting a nilpotent element of G in triangular form it is also easily verified that each nilpotent element of G is ad-nilpotent. Hence a cleaving.in C i s also an ad-cleaving. Thus the notions of cleaving, purity, and nilpotence coincide with those of ad-cleaving, adpurity, and ad-nilpotence, respectively. (Note that ad-cleaving is unique because G is centerfree.) Next assume that G sub (End V),, that is, all elements of G have trace 0. Now observe that an element of G, which is ad-pure as an element of (End V ) , , is also ad-pure as an element of G and the same applies to ad-nilpotence. From this, and the above special case, the same assertions follow. If not all elements of G have trace 0, let G act on a vector space V' spanned by Vand an element w not in V, by g(v

+ aw) = gv - a tr g . w,

for v

E

V, g

E

G,

a

a complex number.

It is easily seen that this yields a faithful representation of G which does not alter the notions of purity and nilpotence; the theorem also follows in this case. 19.20. Theorem Let G E Alg Lie Com Lin admit cleaving, and let A be a maximal pure subspace of G, with H as its idealizer. Then: 19.20.1 H i s a trunk of G, 19.20.2 A consists of the pure elements of H. 19.20.3 H = A + N is a direct sum of Lie algebras in which N consists of the nilpotent elements of H. 19.20.4 N c rad G, even +(N,G) = (0). Proof By the same device used in the proof of Theorem 19.19 one may

suppose that G c (End V),. By 19.19 one may substitute ad-cleaving for cleaving, and so on, ad-cleaving, ad-pure and ad-nilpotence being meant in the sense of (End V),.

98

13-19.

SOLVABILITY AND SEMISIMPLICITY

As i n 17.4, G splits into subspaces G, # (0) with respect to the adjoint image of A , but now (i acts on G, as a scalar multiplication by .(a) because of its purity; the a’s are elements of the linear dual of A . Obviously

[A, Go] = {O},

A

= Go = H .

The ad-pure elements of Go, like all elements of Go, commute with every element of A , hence are in A because of the maximality of A ; this proves 19.20.2. The commutator of an element of A in which all a are different from 0, with an element not in Go,still is not in Go.This implies that H i s its own idealizer in G. The last assertion of Proposition 18.1.1 applied with S = T = H implies that the ad-cleaving components of an element of H , being also in G, are even in H . Since A is in the center of H i t follows that H has rank 0, and H is a trunk of G. Then H = A N as asserted in 19.20.3, since the adnilpotent elements of the solvable Lie algebra H form an ideal of H . Obviously [G,,Gp] c G,+p or = (0). From this it easily follows that:

+

#(G,,Gp) = (0)

if

a

+

# 0.

Putting the image of H in the adjoint of Gin triangular form, one sees that $(H, N> = (01. Hence, $(G, N ) = (0).

So by 19.6 Nis a solvable ideal of G, from which N c rad G.

.

19.21 Theorem If G E Alg Lie Com Lin admits cleaving, the maximal pure subspaces of G are equivalent under Int G, that is, conjugate under G, where G is the group infinitesimally generated by G. Proof Indeed, trunks of G are conjugate under G, and the elements of (7 preserve purity; the assertion follows from 19.20.

19.22. Theorem If G E Alg Lie Com S S , the trunks of G are the maximal ad-pure subspaces of G ; in particular, they are abelian. Proof The assertion follows from 18.3 and 19.20 applied to the adjoint

action. 19.23. Historical Note Theorems 19.19-21 are due to H. de Vries, though partly anticipated by S . Tag6 [Math. Z . 75, 305-324 (1961); J . Sci. Hiroshima Uniu. A-I, 25,63-93 (1961)l. They are closely connected to results of C . Chevalley ( T h h i e des groupes de Lie, Vol. 111, p. 230, 1955).

20-27 DRESSINGS AND CLASSIFICATION OF SEMISIMPLE COMPLEX LIE ALGEBRAS

20. THE FIRST DRESSING OF COMPLEX SEMISIMPLE LIE ALGEBRAS

C E Alg Lie Com SS, dim C = r , t+h is the Killing form of C, and H is a trunk of C . The Caof 17.13.2 are defined with respect to H . 2 0 . 1 4 . The Trunk 20.1 From 17.13.2 it follows that #(x, y)

=

tr fjj = 0

for

Exploiting the first criterion of semisimplicity one gets for any h E H, h # 0 the existence of h , E H with

and for any a 20.1.2

E

W , a # 0 ,f + E C,, f + # 0 the existence of an f -

E

G-, with

t+hcf+,f-) # 0.

20.2 If [ f + , f - ]= 0, then f + ,f- generate an abelian, hence solvable algebra;

takingf+,f- simultaneously triangular and noting that all eigenvalues of f+,fvanish (because of f+C,C Cfi+,) one gets t+h(f+,f-) = trf+f- = 0, which contradicts 20.1.2. Therefore in the case of 20.1.2 one may conclude that [ f + , f -I # 0. Thus, 20.2.1

[Ga,C-al # (0)

for a

In other words, there are a-nodes for a # 0. 99

E

W , a # 0.

100

20-27.

DRESSINGS AND CLASSIFICATION

20.3 The solvability of H can be used to put angular form. Then

h,k, simultaneously into tri-

Hence by 20.1.1 : If L E Hand p(h) = 0 for all p

20.3.1

E

W, then tr = 0.

This proves again that H has no ad-nilpotents # 0. Consequently, by 18.4: 20.3.2

All elements of H are ad-pure.

20.4 If L E H is linearly spanned by u-nodes, 17.15 can be applied which means that p(b) = 0 as soon as u(L) = 0. Then 20.3.1 becomes: 20.4.1

If L is spanned by u-nodes and u(L) = 0, then L = 0.

Applied to u = 0, this yields that every O-node vanishes; thus it is shown anew (cf. 19.22) that 20.4.2

H i s abelian.

If, however, tc # 0, then by 20.4.1 the homomorphism u maps the linear space spanned by [G,, G-,] in such a manner that the kernel of the mapping is (0). Thus its dimension is 0. Hence 20.4.3

dim[(;,, G-,]

=1

if u # 0.

In other words, the a-node (u # 0) is unique up to a scalar factor. Let ha be some u-node. Since L, # 0,20.4.1 shows 20.4.4

u(L,)

#O

for u # O .

By 20.3.1 the number of linearly independent members of W cannot fall short of dim H. Combined with 17.14 this shows that 20.4.5

dim H = rank G.

20.5 Ladders

Choose f+ E G, (u # 0) as a simultaneous eigenvector of the representation of H.Thus

Y h h-

20. 20.5.1

Choosef- E G-, such that yh(f+,f-) an a-node, denoted by h, :

Choose p # 0, p

for all h E H .

[ h , f + ]= a(h)f+

# 0 (see 20.1.2). Thus [ f + , f - ](see 20.2) is

[f+,f-I

20.5.2 E

W, and x

20.5.3

E

101

THE FIRST DRESSING

= h,

# 0.

G,,, x # 0 such that for all h E H

hx=p(h)x

and

f-x

20.5.4

= 0.

(Note that ifp - a .$ Wthen 20.5.4 is a mere consequence of x if p is chosen to imply p - a $ W, 20.5.4 is fulfilled.) An “a-ladder” xo,xI, , .. is defined by 20.5.5

xj

=3;’x,

E

G,,; therefore,

j integer > 0.

Then x, E G,,+ja or = 0. By induction it is verified anew that 20.5.6

hax, = ( p + j a ) ( h z ) x j : hax,+, = t f + x , = . f + h a x j+ [ha,f+Ix, = (P

+ / a ) (ha)x j + 1 + a(ha)XJ+ I

where 20.5.5,20.5.1, and 20.5.6 (inductively) have been used. By an analogous induction,

f-x,

20.5.7

= P,-1 XJ-1

with certain scalar p j : JI-Xj+i

=f-f+x j =f+f-x, - [f+,f-I xj

= Pi-1

.f+xi-1

= PJ-I

x, - ( p + j a ) ( k ) x , ,

- ha x j

where 20.5.5,20.5.7 (inductively), 20.5.2, and 20.5.6 have been used, the basis of the induction being 20.5.4, provided x - ~is understood as 0 and p-l as 0. Moreover, 20.5.8

PI

=Pj-1-

(P +ja)(ha)*

Adding up overj = 0, 1, ..., i, it turns out that

102

20-27.

DRESSINGS AND CLASSIFICATION

20.5.9

Since the number of spaces G, is finite, there is a p such that

xp f 0, X p + , = 0. This makes pp = 0; hence P

C

J =o

(P +1.i (ha) = 0,

(P + 1) Aha) + M P+ 1) .(ha)

= 0,

20.5.10

is the length of the ladder, counting intervals rather than rungs. Again one notes that p(ha) is a rational multiple of a(ha) (see 17.15). The midpoint of the corresponding rootform sequence is 20.5.11

if X=p++a

with an arbitrary real scalar f . Thus all sequences of rootforms corresponding to a-ladders and lying on the straight line p fa ( f variable) have the same midpoint. Rewriting 20.5.9by means of 20.5.10 one finds

+

= +(i

+ l)(p - i ) a(ha).

Hence by 20.5.7 20.5.12 f - f + x , = + ( j + l ) ( p - j ) a ( h , ) ~ ~ # O for j = O , l ,

+

..., p - 1 ,

which can be memorized by noting that j 1 is the rank number of x, in its a-ladder and p - j is the number of elements above x, in its ladder. Thus by inverting an a-ladder one gets (up to scalar factors # 0) a (-a)ladder.

20.

103

THE FIRST DRESSING

20.6-7. Branches 20.6 It is now proved that dim G, = I for a # 0. f+, f-,h, are defined as in 20.5. Suppose that dim G, > 1. Then, since 6, is pure (20.3.2), there is a y E G,, not linearly dependent on f+, with

-

20.6.1

hay

=

a(ha)Y.

Sincef-y is an a-node, it is some multiple of ha because of 20.4.3. Therefore in x = y + uf+ the scalar u may be chosen to make

f-x = 0.

20.6.2

The conditions in 20.5.1-4 are fulfilled with a instead o f p . Hence, according to 20.5.10, the a-ladder initiated by x should have length -2, which is not possible. The supposition dim G, > 1 must be wrong. This proves that 20.6.3

dimG,= 1

for a#O.

20.7. Definition The elements # 0 of G, for u # 0 are called a-branches. Later on they are indicated by e,. It has been proved that the a-branch is uniquely determined up to a scalar factor. 20.8-1 2. Ladders of Rootforms 20.8. Definition For a E W, a # 0, the po, ..., p,, form an a-ladder of rootforms if p,+, - p, = a and forf, E G,, 20.8.1 20.8.2

f+G,,,#{O}

for j = O ,..., p - 1 .

Let p E W and x E Gp. Then y =f_4x # 0, f-f_q+'x= 0 for some positive integer q. By 20.5.11-12, f+qy# 0. So p belongs to an a-ladder which by 20.5.1 I , 20.6.3 is unique. By 20.5.7 the inverse of an a-ladder is a (-a)-ladder. By 20.5.1 1 it happens that all (&a)-ladders on the real straight line p + (a (5 variable) have the same midpoint. Therefore the a-ladder po, . . ., pp contains all p E W with p = po mod a. 20.8.3

f+GG,#{O}

if f + E G , ,

f+#O,

20.9 If po, . . ., pp is an a-ladder, then by 20.5.10

a#0,

a,p,p+aeW.

20-27.

104

DRESSINGSAND CLASSIFICATION

For pf = po +j a

Since any rootform is a member of some a-ladder, this shows that 20.9.1

-2 P(hJ -- is an integer for any p a(h,)

E

W.

Defining

one gets Sm Pi = p.p-j*

This shows that S, manages to invert the a-ladders. Since every rootform belongs to some a-ladder, S, maps W onto W. The definition of S, can be extended. The rootforms are special linear functions on H . One is led to define S , for all linear functions on H . 20.10. Definition H* is the dual space of H (i.e., the linear space of complex linear functions on H ) . For a E W, a # 0, S, is the linear mapping of H * onto itself, defined by

It was shown that S , maps W onto itself; S, leaves the subspace o f f with = 0 invariant, and it changes a into -a. Thus it is a kind of rejection. In any case, SZ2= 1.

[(h,)

20.11

If A, p E Wand p = t A # O with a scalar t , then t = k 1. This is shown as follows: One may suppose It1 2 1.

are integers by 20.9.1. Hence t = +I or +2. In the case t = *2 a A-ladder is formed by -2A, -A, 0, A, 2A. With f+ E GA,xo E G-2A,no# 0, xi =$! xo # 0 for j = 0, 1, 2, 3,4, one would get xo E G-2~9 xi E G-A,

~2 E

Go, x3

E

GA, x4

E

G~A,

where x3 would be a scalar multiple off,; therefore x4 =$+x3 = 0, which is a contradiction. Thus t = *1. Note that all ladders through 0 have the form -a, 0, a.

20.

105

THE FIRST DRESSING

20.12 By means of 20.5.12 the trace off-f+ in C;=oG,,+,, can be computed as the sum of the p j ; thus

+ -En- I

I=O

( j + 1) (P - A a(ha),

By elementary algebra (e.g., induction on p) this turns out to be

-?~P(P + I)(P + 2)a(ha)Hence

where the sum runs over the set of all a-ladders and p is their respective length. 20.13. First Dressing The results are collected in a definition and a theorem. Definition G E Alg Lie Com S S is said to be infirsf dressing with respect to a trunk H if the branches ea and the nodes ha have been normed such that [e,,e-,l = ha. First Dressing Theorem on the complex semisimple Lie algebra G of rank I with the trunk H, its dual H*, the Killing form i,h, and the rootform system W. H is abelian, dim H = 1,

the rootform 0 is I-fold, and all others are simple. Among the rootforms there are I independent and no more. If a E W , then -a E W, and no other multiple # 0 of a belongs to W. To every a E W , a # 0, there is a branch em,unique up to a scalar factor, such that [h,eal= a(h)ea for h~ H, [e,, e-,l ha E H, [e,,epl=Na,peatp#O

=o

if a + / 3 # 0 , if a + / l $ W .

a+/3E

W,

The nodes h, span H linearly; G is spanned by H a n d the branches earwhich are a basis of a linear complement of H. a(!(ha)#O

for a

E

W,

-2[h(h,)/a(ha)]is a n integer for a,h

a#O. E

W, a # 0.

I06

20-27.

DRESSINGS AND CLASSIFICATION

Sa (see 20.10) maps H * onto H*, W onto W, and inverts the a-ladders of rootforms. I f a,h E W, a # 0, then any p E H * which is congruent to h mod a and is situated between h and Sax, belongs to W . If a,h E W , A - a 4 W, any rootform that is congruent to h mod a belongs to the a-ladder from h to Sah. The length of the ladder is -2[h(ha)/a[ha)]. t,h is nondegenerate on G, and on H. For h,h’ E H : $(h,h’) = CPEw p(h)p(h’). For h E H : t,h(h,e,) = 0. For a,p E W , a # 0, p # 0,a p # 0:$(e,,eb) = 0.

+

t,h(ea,e - a ) = -1% I;P ( P + I ) ( P + 2) a(ha) = Nay the sum running over all a-ladders,p being their respective length. If in its a-ladder y is the ( j 1)th element withp -jelements following, then

+

2-a 2, e,, = t ( j + 1) (P -1) 4ha) ey’ where for y = 0 the role of e,, is played by ha.

Definition Na and Na,s are defined by their occurrence in the first dressing theorem. Moreover, Na,s = 0 if a,p E W , a,p # 0, a fi 4 W.

+

20.14. Historical Note The preceding analysis of semisimple Lie algebras and most of the notions on which it depends are due to W. Killing and E. Cartan (see 26.25). Killing’s work exhibits some gaps and errors, yet by no means enough to justify the tradition, going back to S . Lie himself, of belittling or ignoring Killing’s part. 20.15-1 6.t Semisimple Subalgebras 20.15. Proposition Let G E Alg Lie Com S S , let H be a trunk of G, and let F sub G, H c F.Then F is linearly spanned by H a n d some branches of G with respect to H.If, moreover, ea E F implies caE F (a E W ) , then F is the direct sum of a subalgebra HI of H and of a semisimple F l . Proof H being ad-pure, G as a linear space splits under the h ( h E H ) into H

and uniquely determined one-dimensional linear subspaces, each spanned by a branch. The splitting of F is part of the splitting of G, which proves the first assertion. Further, let V be the set of CL E W with ea E F, let HI be the intersection of the kernels of the a E V , and let F Ibe the linear span of the ea and ha with a E V. Then F1sub F, [H,,F,] = (0) and H, I;, = F. To show that F, is semisimple, think of I,/J~,as being degenerate. Then t,hF,(ea,e-a) = 0 for some a E V , from which it would follow that t,hFI(ea, F,)= t,hFl(Fl,e-a)= (0); thus ea,e-, E rad F,.This, however, contradicts ea,e-,, ha spanning a semisimple subalgebra of F,.

+

.t The contents of 20.15-16 will not be used for some time.

21.

FIRST WEYL NORMINC AND SECOND DRESSING

107

20.16 Remark If F is a subalgebra of the real Lie algebra obtained from C E Alg Lie Com SS by waiving, and F contains a trunk of C,then F is even a (complex) subalgebra of C. 21. THE FIRST WEYL NORMING AND THE SECOND DRESSING OF COMPLEX SEMISIMPLE LIE ALGEBRAS

G E Alg Lie Corn SS. The notation is that of 20.13.

21 .I. Norming the Branches By applying the G-invariance of (19.5.1) with the derivation 0-, to #(h,,e,) one gets #G-,hA, e,)

#

+ # ( ~ A A , , e , J= 0.

Since 0-,,h~ = - PA,e-,I 0-, e,, = - [e,, e-,I

e-,,

= =

-h,,

this becomes

21 .I .I

N,

=

#(hh h,J.

Replacing the e, by appropriate scalar multiples, one can make

21.1.2

N,

=

#(e%,e-,)

I.

:

This is called a first Weyl norming, which in the sequel is assumed to have been carried out. Note that the nodes no longer depend on the choice of norming as long as the norming has been carried out according to 21.1.2. Now 21.1.1 reads

21 .I .3

= #(hAJ,).

Consider, for p fixed, the linear function on H

Y h(p(h) - #(h* It vanishes for all nodes hA,but since they span H i t vanishes on all H :

21 .I .4

p(h) = #(h,h,)

for h E H .

This suggests the following. Definition of the (canonical) mapping

5 of H o n t o H * :

5 = Y h’ Y h #(h, h’), in other words

I08

20-27.

21.1.5

(&W))(h) = #(h, br)

DRESSINGSAND CLASSIFICATION

for h, h’ E H.

5 is linear and one-to-one because of the nondegeneracy of #. 21.1.6

5h,

=p

because of 21.1.3. # can be interpreted as an inner product on H. It induces an inner product on H*, written (..., ...) : Definition

(5k5h’)= #(A, h’).

21.1.7

Note that the definition of the inner product on H* does not depend on the norming. With the use of this inner product, 21.1.3 reads 21.1.8 pL(hA)= By means of 21.1.8 formulas like those of the ladders’ length and of S, can be rewritten.

The inner product on H* like that on His nondegenerate, but since H* is a complex space there is no question of definiteness of the inner product. By the first dressing theorem #(hA,

= ffEW

a(hA)

which now becomes 21.1.9

a remarkable formula, which remains true if A and p are replaced by arbitrary linear combinations of rootforms and thus by arbitrary elements of H*. From N, = I (and the formula for N, = #(e,,e-,) in 20.13) it follows that all ( a , a ) ( a E W ) are rational. Since all -2[(P,a)/(a,a)] ( a ,/3 E W, a # 0 ) are integers, it follows that 21.1.10

all ( a , @ )are rational

( a , @E W ) .

On a basis of H* consisting of rootforms p,, ..., p, every rootform p is presented as P=

z rj PI.

21.

FIRST WEYL NORMING A N D SECOND DRESSING

109

From ( P ~ P ,= ) C rj(Pj,Pi) J

the r,, if computed by Cramer’s rule, turn out to be rational. Therefore all rootforms are rational-linearly dependent on I among them. Correspondingly, by 21 .I .6 the nodes are rational-linearly dependent on I among them. 21.2. The Standard Trunk Definition After norming N, = 1, H,, resp. H: is the real linear space spanned by the nodes h,, resp. by the rootforms. They do no depend on the choice of norming N , = 1 ; H,, is called the standard trunk belonging to H for reasons made clear later on. Obviously CH,, = H:, dim H,, = dim H: = 1. H,H* are the complex extensions of H,,,H:. H,, can also be characterized as the maximal subset of H on which all rootforms are real. There is a nondegenerate inner product on H: by restriction. By 21.1.10 it is real-valued. By 21.1.9

which shows the positive definiteness of the inner product on H:. Correspondingly, the inner product on H,, by restriction of $(. . ., . ..) is positive definite. On any linear space spanned by real linear combinations of nodes (over any subfield of Com), $ is nondegenerate. 21.3. The Reflections Thanks to 20.10 S, is a kind of reyection; because

S, a = -a, Sat= ( for ((,a) =0, S, is even orthogonal with respect to the inner product on H*.

S, maps W onto itself and consequently H: onto itself. Its restriction to H: (also called S,) turns out to be the orthogonal reflection in the orthoplement of a with respect to the inner product on H:.

110

20-27. DRESSINGSAND CLASSIFICATION

By means of (-' the reflection S , can be transferred to Hst.It is again called S, (instead of (-I S, 5). In H,, it is the reflection in the kernel of a. 21.4. Second Dressing The results are collected in a definition and a theorem : Definition

G E Alg Lie Corn SS in first dressing is said to be in second

dressing if the branches have been normed according to thejirst Weyl norming, that is, such that Na = $(ea,e - J = 1 I

and consequently the nodes such that p(hA) = $(hA, h p )

for A, p

E

w, A, p # 0,

and an inner product (. . ., . ..)has been introduced in H* such that ( k p ) = $(hA,hp)

for A,p

E

w, A, p # 0.

Second dressing is always possible. Second Dressing Theorem on the complex semisimple Lie algebra G

(see 20.13). Under second dressing one obtains for A,p p(hd = (A,

E

W, A,p # 0,

II.1 = $(h,h p ) = aeW c ( k a )(p,a),

and, more generally, (5,q) =

c (5,4(q,a)

aEW

for 5,q

E

H*.

Furthermore, (A,p) rational

(A, A)

'0

for A,p E W, for

h E W, A#0.

On any basis of rootforms all rootforms have rational coordinates. The second dressing nodes and the inner product on H* do not depend on the choice of the norming. On the real linear space H: spanned by W the inner product is real-valued and positive definite; the same is true of $ on the real linear space H,, spanned by the nodes ha. On any linear space spanned by the nodes (over any subfield of Com) $ is nondegenerate. S, is a reflection in the sense of the inner product.

22. G DETERMINED BY

w*

111

Notation If by means of 5 endomorphisms (and semiendomorphisms) of H and H,, are carried to H * and H:, respectively, and conversely, they are indicated by the same symbols. An immediate consequence is the following: Proposition If T is an endomorphism of H and T’ is its transpose with respect to #, then 5(TW = ( T ’ t ) ( h ) ;

in particular, if T is orthogonal with respect to #, E(TN = W-’

O(4.

21.5. Gordon Brown’s Formula A remarkable conclusion derived from 21.1.9 is (= rank G ) . ( a , a )= I UEW

Proof Considering the symmetric matrix M of (X,p) (h,p E W , A,p # 0),

one can read 2 1.1.9 as M* =M. Clearly rank M = 1. Thus ZuEw (a,a ) = tr M potent matrices trace and rank are equal.

=

rank M

=

1, since for idem-

21.6. Historical Note The norming N , = 1 (actually -1) was introduced by H. Weyl in Mark Z . 24 (1926); see Selecta, 338-342 (1956). Notwithstanding its great theoretical importance, it suffers from the large denominators in the rational numbers #(h,,hp) which make it less practicable for computations in Lie algebras. This is the reason for not identifying H a n d H * in this exposition, contrary to common usage. Gordon Brown’s formula is found in Proc. Amer. Math. SOC. 15, 518 (1964).

22. G DETERMINED BY W *

G E Alg Lie Corn S S in second dressing. The notations are those of Sections 20-2 1 . 22.1 By 17.8 the adjoint group Int G acts transitively on the set of trunks. By this action the W ( C , H )of different H a r e related. The structure of W does not depend on the choice of H.

112 Definition

20-27. DRESSINGS AND CLASSIFICATION

W* is the subset of W consisting of the rootforms # 0. W *

possesses : 22.1.1 the strong structure of a system of vectors in an I-dimensional linear space with nondegenerate inner product, and 22.1.2 the weak structure of a system of things a, /?, y, abstract relations a /3 = 0 and a fi y = 0.

+

+ +

..., related

by

22.1.3. Proposition The strong structure of W* is fixed by the weak structure. Proof Given the weak structure, one can subsequently determine the ladders,

their lengths, the mappings S, as far as they act on W * (namely, inverting the ladders), the (a,a) by formula 20.13 for N, = #(ea,e-,) = 1, the (a,/3) by the ladder length formula, all linear relations among the rootforms by the values of (a,/3) because of the nondegeneracy of the inner product, the rank as the number of independent ones, and the embedding of W* in H: up to orthogonal mappings. 22.1.4 The influence of the choice of H on the structure of W ( G ) was eliminated by the adjoint group. This appeal to Int G can be avoided. With a continuous change of H, the branches, after suitable norming, change continuously. So do the $(earre-,),and therefore the norming factors needed for second dressing may be assumed to be continuous. Then the second dressing nodes depend continuously on H. Since, by the end of 21.1 they are rationallinearly dependent on I among them, the rational-linear, hence the real relations among the nodes and, consequently, among the rootforms are invariant under a continuous change of H. This shows anew that the weak, and consequently the strong, structure W ( G ,H) does not depend on H. 22.2. Theorem W* determines G up to isomorphy. For two complex semisimple Lie algebras G, G’, with trunks H, H ’ , any isomorphism of W*(G,H ) to W*(G’,H‘)is induced by some isomorphism of G onto G ‘ , thus relating H to H’ and mapping nodes into nodes and branches into branches after suitable norming in second dressing. The theorem is proved in 22.3-5. 22.3 W*(G,H) and W*(G’,H’)are identified according to the given isomorphisms; notation W*. G,G’ are assumed in second dressing with respect to H , H ’ . Thus N,(G, H ) = N,(G’, H ‘ ) = 1.

22. G

DETERMINED BY

w*

113

The normed a-branches and a-nodes are called ex, e;, h,, h:. hi is made to correspond to h,. This correspondence is extended by 21.1.6 to a linear mapping of Honto H ’ . A scalar multiple of e: is to be assigned to e, as its image. Independent of the choice of this multiple, the relation [h,e,l

=

0 )e,

( h E HI,

remains valid if h, e, are replaced by their images. Rather than mapping e, into a multiple of e:, it is advisable to renorm the e;. In doing so, one has to keep N,(G‘, H ’ ) = 1. Thus renorming means multiplying e:, e’, by reciprocal factors. Then t,b(e;,eL,), as well as h; = [e;,el,], is preserved. To settle the isomorphy of Gand C ’ ,one has to renorm thee; such that [e:,

41

=

[ear,91’

(Of course, for any scalar K , ( K e A ) ’ means K e i . ) 22.4 Suppose a,/3,y E W*, a

+ /3 + y = 0. If

[e;, ebl= [e,, epl’,

then [ei,e;I

=

[eg,eyl’.

Proof In

“e,, egl, eyl + “eg, eyl, eul + “ey, eul, 91 = 0

and

“4,eil, ell + “eb, e;l, 41 + “e;, 41,$1

=0

the first summands are the same multiples of hy and h;, respectively, the other summands are multiples of h,, hg and hL,hb, respectively.Thereisessentiallyone relation between them: hy ha hg = 0,

+ + h; + h; + hi = 0

(see 21.1.6 and 20.11). Thus the last two summands are the same multiples of h,,hp respectively hi,/$, and consequently [ea,ey], [ei,e;] are the same multiples of e-, and L,. This proves the assertion. 22.5 On an ordered basis of H,*, the elements of W are lexicographically ordered with respect to their coordinates. The order relation is denoted by >. For p > 0, p E W * the set of all p E W* such that -p < p < p is called W p . Suppose that renorming of the e,’, has been achieved within W p :

114

20-27.

22.5.1

DRESSINGS AND CLASSIFICATION

[ei,e;] = [eA,e,]’

It is expanded to

for X,p, X

+ p E W,.

w,u {PI ” {-PI.

This means that by renorming ei,eL, one has to satisfy

22.5.4

[ei,eL,] = [e,,e-,]’.

+

If p # X p for all h,p E W,, then the same is true of -p, and h + p Then 22.5.2-3 are void, whereas 22.5.4 is fulfilled in advance. Thus suppose for some a, j3 E W,. p = u j3

4 W,.

+

e i is some multiple of [ek,el;].It is renormed by requiring that it be the same multiple of [e;,el;] as e, is of [e,,ep]. Then

is fulfilled. Likewise one can make 22.5.6

[eL,, eLp] = [e-,, e-p]’.

The norming N p = 1, however, might now be lost. Moreover, the definition of e;, seemingly depends on the choice of u,/3 in p = a + j3. Let y,6, with p = y + 6, also be a choice (y,S E W,). It must be shown that 22.5.5 remains true with y , 6 instead of a, j3. (This implies that e;, d o not depend on the choice of Q,

B.> By Jacobi-associativity

22.5.7

“4,eil, e’,l + “e;, eL,l, 41 + “eLy,41,$1

+

=

0.

+

Now j3 < p, hence p = a /3 < a p, hence 0 < a ; thus 0 < u < p. Likewise 0 < j3 < p, 0 < y < p, 0 < 6 < p, hence -p < a - y < p,

-p < j3 - Y < p ;

if a - y, /3 - y are rootforms, they belong to W,. Finally, 6 = u + /3 - y E W,. Consequently the second and third summands of 22.5.7 are constructed within W,. There one may pull the dashes out of the brackets. But then it is also allowed in the first summand

23.

SECOND WEYL NORMINC AND THIRD DRESSING

115

22.5.8 This permission extends to the case y

= a,

hence to the expression

and likewise to “[e;, e i l , e’bl, e’al, thus by Jacobi-associativity to “ek, eb19 [e’a e’fill, and by 22.5.6 to [Lea,ebl‘, [e-a, e-fil’l, and consequently to Lei, e’, I . Now 22.5.4 is satisfied and so is the norming N p = 1. It is possible to apply 22.4. From 22.5.8 it follows that [e;, el_,] = [e,, e-yl’,

thus half of 22.5.3. Thanks to 22.4, this proves the validity of [e’,, eL1 = k ye-d’, ,

which is half of 22.5.2. The other halves are obtained by interchanging p and

-P. 23. THE SECOND WEYL NORMING AND THE THIRD DRESSING OF COMPLEX SEMISIMPLE LIE ALGEBRAS

G E Alg Lie Com S S in second dressing. The notations are those of 20.13.

23.1 The mapping a + -a is an automorphism of W*. According to 22.2, it is induced by an automorphism M of G with Mh,

=

h-,,

hence Mh = -h Me,

for h E H ,

= v,e-,

with scalar v,. Because of Mh,

=

M [ e , , e - , ] = [ M e , , M e - , ] = v,v-,[e-,,e,]=

one obtains

23.1.I

VaV-,

=

1

v,v-,h-,,

116

20-27. DRESSINGS AND CLASSIFICATION

For all a E W*, e, is replaced by e:

=dCaea,

where because of 23.1.1 the square roots can be chosen to arrive at

d< dTa= 1. This norming retains N, = 1. 23.1.2

Me:

=

d r , Me, = G

v , e-,

=

d<e-,

= e:,.

The process just carried out is called second Weyl norming. It brings G into its third dressing. More precisely: Definition An automorphism of G E Alg Lie Com SS is called a minusautomorphism with respect to a trunk H (and generally indicated by M) if it preserves Hand behaves on H a s the scalar multiplier -1. Note that M is not uniquely determined by H.Any element of (exp l?)M satisfies as well. The converse holds too (see 33.9). Proposition

M 2= 1.

This follows from 23.1.2. Definition G E Alg Lie Com SS is said to be in third dressing if a pair ‘H, M l has been chosen consisting of a trunk H of G in second dressing

and a minus-automorphism M with respect to H,which are connected by the requirement Me, = e-, (a E W,a # 0). Third Dressing Theorem Third dressing is always possible, namely by a

readjustment of the branches which preserves second dressing. Note that this readjustment depends on the choice of M , but even if M were fixed the ambiguity of the d c would allow for changing ‘e,,e-,l into ‘-e,,-e-,l for any set of u E W * . This, however, is the only indeterminacy left. 23.2 Assume that G is in third dressing. The structure of G is settled by the knowledge of the N,,s in

[ea,efiI=Na,fiea+B

(B#--oL)*

24.

THE UNITARY AND STANDARD RESTRICTIONS

117

By applying M, one gets [e-,,e-pI

= N,,pe-,-p;

hence 23.2.1

N,.p

= N-,*-p.

According to 20.5.12, L % e p =%+ N p - j ) ( ~ 9 e p if j3 is the ( j

+ 1)th element in its cr-ladder of lengthp. Now LZffep= N-ffsff+pNff,pep9

hence N-ff,ff+pN,*p

=Ki+l)(P -d(%4*

By Jacobi-associativityapplied to e-=, e q , N-,.-pL-p

one obtains

+ N-p.ff+p4z+ N,+p,-ffAp = 0.

Thus, because of R - q - p + R f f + h - 0 and the independence of R,,bp for p. j3 # -a, a ( # 0), all N in that equation are equal. Hence

and by 23.2.1 N,.p

=

W ( H j + l)(P -A(a,a))*

This proves the following : Theorem In third dressing, if j3 is the ( j + 1)th element in its a-ladder (of length PI, N,.p = k i d ( W + 1)(P 4)= N-ff.-p,

-.x%

thus purely imaginary. 23.3 In third dressing the N,,p are uniquely determined up to factors f l .

24. THE UNITARY AND STANDARD RESTRICTIONS OF A SEMISIMPLE LIE ALGEBRA G E Alg Lie Corn SS in third dressing. The notation is the same as that preceding.

118

20-27. DRESSINGS A N D CLASSIFICATION

24.1-4. Standard and Unitary Semimorphism 24.1. Definition The semilinear (see 1.2) mappings C,, and C,, of G in

third dressing onto itself are defined by C,, h = h

C,,

for h E H,,, C,,ea = -ea. ;

= MC,, = C,, M

hence C,,h

= -h

for h E H,,, Cunea= -e-a.

Proposition C,, and C,, are involutory semimorphisms (see 1.6, 1.11) of G. Proof It suffices to show that

24.2. Definition C,, and C,, are called the standard and the unitary semimorphisms of G in third dressing (with respect to H ) ; the Csl- and C,,restrictions of G (see 1.6) are called the standard and unitary restrictions: G,,

and Gun. The standard and unitary trunks (in H ) are H,, = H n G,, and H,, = H n G,,, and I),,, I),,, are the restrictions of I) to G,,, Gun.Their signatures are called the signatures of GsI,G,,, respectively. If G is given as a linear Lie algebra, then G,,, G,,, H,,, H,, are the linear Lie groups infinitesimallygenerated by G,,, G,,, H,,, Hun. More generally, if C is an involutory semimorphism of G and H is a Cinvariant trunk of G, then H n G, is called a C-trunk of G or a trunk of G,. The signature of G, is the signature of $., By Definition 1.6 the C-restriction consists of the u = Cu or, equivalently, the u + Cu. This means that

24.

THE UNITARY AND STANDARD RESTRICTIONS

119

G,, is the real linear space spanned by H,, and the ie,, Gunis the set of iR + CaEW* Toreawith R E H,, and T , + = 0 (? the conjugate of T ) . Hun = iHst. Note that H,, and Hundo not depend on the choice of M and that Cst,C,,, C,,, Gundepend on Hand M only (not on the choice of the signs in the ea). In 33.13 it is shown that a unitary semimorphism is a unitary semimorphism with respect to any trunk left invariant by it. 24.3. Theorem G,,and Gunare real Lie algebras with the complexextension G. The signature of G,, is rank G, that of Gunis -dim G; +,,, is negative definite. Proof The first assertion follows from the general principle of Section 1. +st

is positive definite on H,, (21.2),

&t(kea) = 0 for R E H,t, +st(C T a iea, C T a iea) = -C T a T - a , which shows that signature +,, = dim H,, = rank G.

+,,, is negative definite on Hun= iH,,, +Jilt, +un(

C

(Taea

ea - ?a e-a) = 0, 2 (Taea - T e e - a ) ) z - 2

T~

- tae-a),

a>O

a>o

C

7 a ?a,

a>O

which shows that $un is negative definite. 24.4 Together with an orthonormal basis of Hunfor &,, the elements

2/.)(ea

-

e-a),

i2/$(ea

+ e-a),

with u > 0,

form an orthonormal basis of Gun for -$,,,,; Int (Gun)leaves #,,, invariant. Therefore, being connected, it is a group of rotations and, as such, is bounded. By Theorem 19.10 it is closed, hence compact. Theorem Int Gunis compact.

This fact explains the particular importance of the unitary restriction. In 32.2.4 Gun itself is proved to be compact (if G is linear). The standard restriction arises in numerous geometric contexts. In general, there are more real restrictions than the unitary and standard ones: 24.5. Historical Note Though the unitary restriction was already known in any particular case, H. Weyl first recognized its importance and proved its

120

20-27.

DRESSINGS AND CLASSIFICATION

existence by a method which is essentially that of Sections 22-23 [Math.2. 24 (1926); see Selecta, 342-346 (1956)], though he took its compactness for granted. 24.6-7. Examples

24.6 In 19.14 it was shown that the G E A,, B,, c,,D,

of Section 16 are semisimple. It is worthwhile to find out what standard and unitary restrictions mean i n these examples. It is easily seen by means of 17.7 that the subalgebra H of G indicated in Section 16 as a trunk is a trunk. The branches e,,e-, were chosen with real matrix coefficients and were transposes of each other. They do not obey the first Weyl norming. However, by computing [Lea, e-al, eal

in the particular cases, one can verify that 4 h a ) >0

(this result could be used for another proof of the semisimplicity of C ) .Now, by 20.12.1 it turns out that N , > 0. The first Weyl norming can be performed by multiplying e,,e-, by equal real factors. After this norming e,,e-, are still real and are still transposes of each other; ha is also still real and consequently H,, consists of real diagonal matrices. For matrix groups )‘,,a’-l is an isomorphism. So is 7 = Y ,,(-a‘)

for Lie algebras of matrices. For G E A,, B,, C,, D,, as presented in Section 16, 7 is even an automorphism: in A, because tr a = 0 implies tr(-a’) = 0; in the others because the defining relation a’s’+ su = 0 remains true if a is replaced by -u’ (note that s = s-’). NOW -~#=-h for h E H, Tea = -e-a, which shows (see 23.1) that one may put q=M

24.

THE UNITARY AND STANDARD RESTRICTIONS

121

and that the second Weyl norming can be performed by replacing e,, e-, by ie,, -it?-,, where a runs through a set Xsuch that X n (-X) = 0and X u (-X) = W*. The standard restriction is really spanned by H,, and by the new branches with factors *i, that is, by the old branches. Thus the standard restriction consists of the matrices defined in Section 16, interpreted with real matrix coefficients. For this very reason it was called the standard restriction. The unitary restriction means that opposite branches must have opposite conjugate coefficients. This requirement is not influenced by the factors i, -i at e,,e-,. Further, Gun has iH,, as its trunk, which makes the diagonal coefficients purely imaginary. Since e,,e-, were transposes of each other, the unitary restriction in the matrix representation consists of matrices satisfying u

-

+ u’ = 0.

This equation characterizes the infinitesimal algebra of the group of unitary matrices with determinant 1, which explains the term unitary restriction. The result can be formulated as follows : Theorem For G E A,, B,, C,, D,, as introduced in Section 16, the standard restriction is obtained by restricting the matrix coefficients to real values; the unitary restriction is obtained as the intersection with the infinitesimal algebra of the group of unitary matrices. The latter property is extended in Theorem 38.4.

24.7. Proposition On a suitable basis Gunof G E B,, D,, as defined in Section 16, is the infinitesimal algebra of the ordinary real rotation group in (21 + I)- and 21-space. Proof for D, (the case B, is much the same).

Let e l , ..., e2, be the ordered basis with respect to which the quadratic form is described, that is, has matrix s. Now the unitary restriction of this member of D, consists of those endomorphisms that leave both the quadratic form and the canonical positive definite hermitean form (with matrix 1 on e,, . . ., e2,) infinitesimally invariant. Take as the elements of a new basis j = 1 , ..., I , b,=+(l+i)e,++(l -i)e,+,, b,+,=+(l-i)e,+f(l+i)e,+,, j = 1 , ..., 1. This new basis is orthonormal with respect to both the quadratic and the hermitean form. Hence the matrices on this basis of the elements of the unitary restriction of this member of D, are those that are both skew and hermiteanskew, that is, real and skew. This proves the assertion.

122

20-27. DRESSINGS A N D CLASSIFICATION

25. G DETERMINED BY W++ G E Alg Lie Com SS, if needed in second dressing on a trunk H.

25.1. Angles of Rootforms

The notation remains the same. For independent A,p

E

W* under the angle

c (A,p) (according to the inner product in H:) 25.1.1

is an integer (20.13), 4c0s2 C (A,p) 0;

hence (.,pi) < 0. This proves the existence of a,p (namely, p,), as required. The value (a,p)is determined by the presentation of a as a sum of elements of W++and by the values of (p, u) for p, u E W + + This . settles the remainder of the proposition. 25.4.1. Proposition The coefficients of a E W +on the basis W + +have no common divisor (except 1). Indeed, the assertion is evident for primitive rootforms and is preserved in the building up of W + from W + + since , (in the preceding proof) any common divisor of the coefficients of p reappears in m,hence in the coefficients of a = - mP*

B

25.4.2 An immediate consequence of the construction of W + from W + +is the following: Proposition G is generated by the eP,e-,with p

E

W++.

25.5. Direct Splittings Definition A splitting of W* into mutually orthogonal sets is called direct. A splitting of W++into mutually orthogonal sets or, equivalently, into subsets with no bond from one to another is called direct. Theorem Direct splittings of G, W * , and W + +go together and induce each

other. Proof Let G = GI

+ G2 be a direct sum. If H I ,H2 are trunks of G1,G2,then

H = H , + H2 is a trunk of G, and all trunks of G are obtained this way, thanks to the equivalence of trunks under the adjoint group. Branches of G,,G2commute with each other because [ G l , G 2 ]= (0). Therefore rootforms from GI,GZare orthogonal; this is particularly true of rootforms in W + + ,which consequently splits according to the assertion. Conversely, if W++splits into nonconnected and consequently orthogonal subsets W i + ,rebuilding W* by 25.4 only creates sums for every W:+ separately. They form orthogonal W,?, which constitute W + . Adding the opposite rootforms produces W;”. For a, E W;” ( j = i, k, i # k ) one gets (at, ak) = 0 as well as aI+ dk # W , hence [hai, haJ= [ha,, e a J = [eat, earl

= 0,

which make G a direct sum of the G, corresponding to W;+.

128

20-27.

DRESSINGS AND CLASSIFICATION

Corollary G is simple if and only if

W + +is connected.

25.6. The Top Rootform Proposition G E Alg Lie Com SSS (i.e., simple C ) in ordered dressing has one maximal rootform (i.e., exceeding all other rootforms). Proof Let a and /3 be maximal rootforms, a #

P = Z b,p,

By

W++). Of course, a,/3 E W+, b, > 0. Now a - /3 4 W + because /3 is maximal and /3 - a $ W +because a is maximal. Therefore a - B 4 W,whence (a,/3) = 0, (P,E

r, b,(a, P,)

= 0.

The maximality of a implies (a,p,) 2 0 for allj; thus (a,p,) # 0 + b,

= 0.

Because of a # 0, there is a j with (a,p,) # 0 and consequently with b, = 0. The graph of G is connected. Therefore there is some p E W + +with a neighbor u E W + +and such that p occurs in /3 with a coefficient > 0, whereas u does not. Thus (Byu)< 0, /3 + u E W+, which contradicts the maximality of 8. Definition If there is a unique maximal rootform, it is called the top rootform. Note that the existence of the top rootform is equivalent to the simplicity of G. 25.7-8. Examples 25.7 In Section 16 natural bases of W *were found for G E A,, The corresponding graphs are : dimension AI:

0-C-O

PI

8,:

p,

p3

p3

p*

PI-2

PI

PI-1

1(21+ 1)

*.

**o.*.-c4a PI p1 p3

DI: --o--o p3

( I + 1)’ -1

*-O---O

- *-

**CJ

pa

c,:

*

PI-1

PI

PI

P I 4

PI4

PI

1(21+1)

0-0

p4

PI

PI4

PI

g 1;

l(21-1)

B,, C , , D,.

25. G DETERMINED BY

w++

129

The primitive rootforms have been linearly arranged as in Section 16, A,: ~j = w j - wj+i, B,: p , = w , - , - w ,

for j > l , p , = w , , for j < l , p I = 2 w 1 ,

C,:p,=w,-w,+, D,: p, = w , - ~- wJWl for j z 3,

p,

= wI-l

- w l , p2 = w,-,

+ wI.

It is easily verified that no other bonds than those appearing in the diagrams are possible; for example, w , - w2 and u3- w., are not joined because (0, - w 2 ) + ( w 3 - u4)is not a rootform; w , - w2 and w2 - w 3 are simply joined because ( w , - w 2 ) + (w2 - 9 )= w , - w3 is a rootform, whereas neither (wI - w 2 ) 2(w2 - w 3 ) nor 2(w, - w 2 ) + (w2 - w 3 ) is a rootform. In the case of C E B, there are rootforms ( w I - ,- w , ) + j w l f o r j = 0,1,2; hence

+

-2

(WI-1-

w 1 9

w1) -2

(w1,wl)

and (see 25.1)

-2

- WI,w,)

@I-I (UI-1

- 0 1 , w1-1 -

4= 1 ;

thus (w,,

4< ( U l - 1

- WI, U l - 1 -

4,

which settles the direction of the arrow. For G E C , it points the other way. 25.8 The graphs disclose a few isomorphies: 25.8.1

A,

= B, = C ,

(DI was not defined).

25.8.2 25.8.3

6 2 = C2.

D2 = A,

+ A,

(which is a way of saying that a member of D2is the direct sum of two of A,). 25.8.4

A3

=

D3.

These isomorphies can be proved directly. The proofs are sketched. 25.8.1' By definition, A, = C , is trivial; A, = B, reflects the well-known local isomorphism between the projective group of the complex projective line (function theory sphere) and the rotation group in 3-space by means of the stereographic projection between plane and sphere.

130

20-27.

25.8.4‘ A, by

(a)

=

DRESSINGS AND CLASSIFICATION

D,.Linear mappings with unit determinants in 4-spaceYdefined

5;. = c1 at“ 51,

induce mappings in the 6-space of skew tensors of degree 2: , n*,,*=

- aJ’l

(aI’l

1hJ

aI’J)

This leads to a representation G’ of G E A, in 6-space. The variety of bivectors with coordinates

51

52

53

54

rll

rl2

7?3

74

5,

52

53

54

?)I

112

7?3

7 4

=o.

25.8.2‘ B2= C2.With the same notation, G E C2 is the group of transformations (a) leaving the skew bilinear form

c

UII

5, 7,

on 4-space invariant. In 6-space this implies invariance under G’ of a linear form : uIinlJ

(for bivectors rnl,l, thus for all skew tensors rnl,l by linear combination). The nullspace of this form is a 5-subspace invariant under G‘. Its intersection with

26.

CLASSIFICATIONUP TO ISOMORPHISM

131

(c) is a quadric in 5-space and is again invariant. Consequently G ' belongs to B2.

Inprojectivegeometry?; a i j , $ i ~ j ~ O d e f i n e ~ a s o - c a l l e d c o m pollejxr i; j~= O gives the lines lying on the complex.

+

25.8.3' D, A, A,. The projectivities of 3-space leaving a quadric invariant also leave invariant or interchange the two families of straight lines on the quadric. The second kind may be disregarded. The projectivities leaving every line of one family invariant form a subgroup that is essentially the projective group of a projective line. The whole group is the direct product of the two subgroups belonging to the two families. This verifies the assertion. An alternative argument: the rotations of 4-space can be presented by quaternion multiplications D,.,, 1

D P , , x= p ~ q - '

(lPl = 141 = 1).

The same is true of rotations B, of 3-space:

B,x

= pxp-'

(Re s = 0, I pi

=

1).

The assignment of the pair rBp,Bq' to Dp,qproduces a homomorphism of G E D, onto G ' E FBI, B,1 with the kernel consisting of D l , land D,.-l. It is a local isomorphism that induces an isomorphism of Gand G'.

26. CLASSIFICATION OF SEMISIMPLE COMPLEX LIE ALGEBRAS UP TO ISOMORPHISM 26.1 Because of the second criterion on semisimplicity, C i s supposed simple of rank 2 1. 26.2 All systems W" must be determined. At present the lengths of their elements are not taken into account. The elements are normed as unit vectors. One has to solve the following problem: Problem To find the class B of nonvoid systems P of vectors in Euclidean space such that 26.2.1

ccEP+lorl=l,

26.2.2 the elements of P are linearly independent, 26.2.3 forCL,BEP,a#B,(CL,B)=-~1/S withs=0,1,2,3,

132

20-27. DRESSINGS AND CLASSIFICATION

26.2.4 P is not a union of mutually orthogonal proper subsystems. 26.3 With the notation of 26.2.3, a,p are called s-fold joined if s # 0, unjoined if s = 0. The graph of P is connected because of 26.2.4. 26.4 [PE 8 A P’c P A graph P‘ connected] -+ P’E 8. 26.5 The graph of P

E

8 is not a “circle.”

Proof Suppose that P consists of different a l , j = 1, ..., m mod m. Then

.. ., a,,,,with (a,,a,+l) # 0 for

which cannot happen because of 26.2.2. 26.6 From 26.4-5 it follows that the graph cannot contain a circle. By 26.3 it is a %ee.” 26.7 If a E P is joined to exactly k elements of P,it is called k-sided. If P E 8 and a is k-sided, then P\{a} breaks into k nonvoid subsystems belonging to 8,called the arms of a. 26.8 If P is some set of vectors, L ( P ) denotes the linear space spanned by P. 26.9 For P E B and a E P,define q(P,a) = square of distance a,L(P\{a}).

Then O 0. If it is three-sided, its joins are simple, if it is two-sided, one of its joins is simple and the other may be simple or twofold. 26.12 If P , P , ~ 9 ( i = 1 , 2 ) P , , u P , = P , P l n P 2 = { a } , and v ( P I , a ) < + , then y ( P ,8) G 3 for any j3 E P,. Proof For P2 = { a ,p} this is a consequence of 26.10 because

Induction on the number of steps needed to reach 26.9 complete the proof. 26.13 If a

E

P

E 9is at

p from a and application of

least two-sided, then v(P, a ) G +.

Proof Let a be joined to

a,,a2.

It suffices to prove the assertion for the

134

20-27.

DRESSINGS AND CLASSIFICATION

+

26.14 If /3 E P €9is three-sided, then y(P,a) < for any a

E

P.

Proof Thanks to 26.13, the case a = /3 may be waived. Suppose that a is in the arm P' of p and put P " = P\P.' Then p E P" E 9'. In P", p is at least twosided; hence, by 26.13, v(P",p) < and by 26.12, y(P,a) < +.

+

26.15 P

EB

can have at most one three-sided element.

Proof Assume that a, p are three-sided. Let P j ( j = 1, 2, 3) be the arms of a, and a, E Pi joined to a. Assume that p E P I . Then v ( P j , a j )< 1 and by 26.14 v ( P l , a l )< 4.NOW 1 1 1 v ( P ,a ) < 1 - - - -=0

4.f

4.1

4.1 ~

'

which is not allowed by 26.2.2. 26.1 6 No P

E

9with more than two elements has a threefold join.

Proof

It suffices to refute P = {a,P,y} with a threefold join between a and and a join between /3 and y. This is contained in 26.11.

+

26.17 If P E B has a twofold join, then q(P,y) < for all y Proof

E

p

P.

By 26.12 it suffices to prove 26.17 for P = { a ,p} : 2 y(P, p) < 1 - - = +. 4-1

26.18 No P

E

9possesses both a three-sided element and a twofold join.

Proof Suppose that c( E Pis three-sided and ,f3, y are twofold joined. They may be arranged so that an arm P' of y contains a, p. If a # p, then a is three-sided in P ' ; hence, by 26.14, y ( P ' ,p) G f . If t~ = p, the same follows from 26.1 1. Now by 26.10

2 y ( P ' u { y } , y ) < 1 ---0. 4.3 26.19 No P

E

9has more than one twofold join.

Assume twofold joins between u and fi and between y and 6. They may be arranged so that an arm P' of 6 contains a,p, y. By 26.17 p(P', y ) G f ; hence

Proof

i

26.

135

CLASSIFICATION U P TO ISOMORPHISM

26.20 A P E 9'with a twofold join has a linearly arranged graph according to 26.18. I f P has more than four elements, then the twofold join can stand at one of its ends only. Proof Consider P consisting of different a ] ,a2,a,, m4, 03,with and a2,a3twofold joined. Put P, = { a I , . . ., R j ) .

1 4 ..I.3

a i ,m i + ]

joined

_. I

-

5,

1 4-.L4 =o. 26.21 Let K , ioined. Then

Proof

E

.P be the system of different a , , . . ., CL,, with

c c j , g j + , simply

Obvious for q = 1. Induction

26.22 Let 6 E P E @ be three-sided. By 26.15 and 26.18 the arms of 6 have the form K,,, K,, K, (see 26.21). Hence

From a function table fory = s/(2(.v

+ I)),

s = 1 , 2 , 3 , 4 , 5 , ...)

y

=

$,

4, 4, +,-,5?, . . ., increasing,

it follows forp,q,r that not all of them are > 2, no two of them 2 3. Suppose that p G q G r. Then the only admissible p,q,r are

p =q = I , p=l,q=2,

arbitrary, r ~ 4 .

I'

20-27.

136

DRESSINGS AND CLASSIFICATION

26.23 In addition to the known connected graphs belonging to A, (I > I),

B, ( I

> 3),

C , (I

2),

D, ( I 2 41,

the only possible ones are : dimension (see Table B)

14

Gz: P2

PI

F4: o-----o P2

52 P4

P3

PI

78

Indeed, according to 26.16, the systems with threefold joins are exhausted by G,. According to 26.18, 26.19, and 26.20, the systems with twofold joins are exhausted by B,, C,, F4. According to 26.15 and 26.22 those with threesided elements are exhausted by D,, E,, E,, E,. By these graphs the vector lengths in W++are settled as well. An essential choice of how to point the arrow exists in the cases B,, C , ( I > 2) only. 26.24 Complex Classification Theorem The isomorphism classes of simple semisimple complex Lie algebras are given by A, ( I 2 I),

B, ( I > 3), G,, F4, E,,

C , ( I > 2)s

E,,

D, ( I > 4),

E,.

Actually, the existence of the five exceptional Lie algebras has not been proved here. In 27.1 it is done for G,.(See also 26.25.) Likewise, it has not yet been proved that all of these classes are different. This depends on the proof that the order chosen in H: does not exert any influence on the structure of W + +(see 25.3 and 33.5).

27. G 2 A N D F4.THE CHEVALLEY DRESSING

137

By the procedure in 25.4, W . +can be built up from W + + .This also yields the dimension of the corresponding Lie algebra. The results of these computations are to be found in Table B. The top rootforms are collected in Table D. The construction of W + (F4)is shown in 27.2. 26.25 Historical Note The classification goes back to W. Killing, Math. Ann. 31 (1 888), 33 (1 889), 34 ( I 889), 36 (1 890) and E. Cartan (These, 1894). B. L. van der Waerden simplified it by using the second Weyl norming [Math. Z. 37, 446-462 (1933)l. As a new tool, E. Dynkin introduced the notions of order and of natural basis, and the graph [Uspehi Mat. Nuuk N.S. 2, 59-127 (1947)], though graphs like these had already been used by L. Schlafli and H. S. M. Coxeter in similar situations. The present method was published by H. Freudenthal [Proc. Kon. Akad. Wet. Amsterdam 61,379-383 (1958)l. The exhibited numbering of the primitive rootforms corresponds to Cartan’s of the fundamental weights (see Euures I, 1, 355-398), which is the most natural (except in the case E,, in which it contains an inconsistency). Though it has become a habit for every author to use his own numbering, it would be better to stick to Cartan’s as has been done here. In E. Cartan’s thesis the existence of the simple Lie algebras occurring in the complex classification theorem has been ascertained by individual construction. A general, though cumbersome, construction was devised by Harish-Chandra (Trans. Amer. Math. Soc. 70, 28-96 (1951). Another, still involved, has been designed by J. Tits (unpublished). According to H. de Vries (unpublished), the case of simple bonds only can be dealt with by a rather easy method. The exceptional Lie algebras have been extensively studied in the last 15 years. The literature on this subject can be found in part in the bibliography in H. Freudenthal, “Lie Groups in Geometry,’’ Aduan. Math. 1, 145-190 (1965). A general formula for the construction of all five exceptional Lie algebras has been given by J. Tits (Publ. Math. I.H.E.S. 31,525-562 (1966); see also R. D. Schafer, Proc. Kon. Akad. Wet. Amsterdam A69, 64-69 (1966) (Indagationes 28). 27. G2 AND F4. THE CHEVALLEY DRESSING 27.1-2

G, and

F4

27.1 G E G ~ . (PI,P,)=c,

(P,,PJ=-3c9

with some c yet to be determined. Because of

(pz,p2)=3c,

138

20-27.

DRESSINGS AND CLASSIFICATION

there is a pl-ladder of length 3 from p2.This gives the rootforms PZ

+ PI,

p2 +

p2 +

PI.

Because of

there is still a rootform

2 3/31 ,~ 2 =) 1, -2 ( ~ + (f27f2)

2P2 + 3P,. Furthermore, there are the opposite rootforms and twofold 0. Hence

dim G = 14. P2

- P2 FIG.1.

There are 1 2 1 4

p,-ladder of length 2 p,-ladders of length 3 p,-ladder of length 2 p2-ladders of length 1 Npz =4(t%,P2).

]

(The N p l ,Np2have been computed by 20.13.) Since N p ,= 1, c = i+. To compute N,,p the formula in Theorem 23.2 is used. Then (with E =*I) Np1.p2 = iEl

NpI.pI+pl=

iE2

N p I . z p , + p l = iE3 N p z . 3 p 1 + p 1 = iE4 Np,+p2,2p1+p1

= iE5

4, 4 4

3

4, 4

27. G2 AND F4.THE CHEVALLEY DRESSING with the relation E, E3

by Jacobi-associativity. These

+

€4 € 5 =

139

0

determine the others.

27.2 The method in 25.4 produces the rootforms of G E F4 (Fig. 2). 1000

ol!
1, Now suppose that ho E DD that of E(h,) in E is 2, and consequently, that of the above Can-piecesis > 3. There is a finite number of different D(ho). Hence: Proposition F* is acountable union of Can-piecesof codimension > 3 in G. Remark The constituent Can-piecesmay be assumed such that their closures are still contained in Can-piecesof the same dimension.

158

28-38.

TOPOLOGICAL A N D INTEGRATIONMETHODS

31.9 F* is contained in the boundary E \ F o f open F. Conversely, ifc E E \ F , then c = lim c, with some c, E F, hence with c, = 6,h, (a, E G, h, E D). Because of the compactness of G, one may assume lim a, = a E G, lirn h, = h E b. Then c = Gh E F,, which proves that F \ F c F*; thus

F* = E \ F . According to dimension theory codim F* 2 3. Thus the boundary of F has codimension 2 3. In a manifold the complement of a closed set of codimension 2 2 is connected. Therefore G\(E\F) = F u (G\F) is connected. Since both summands are open and their intersection is void, this means that G\F is void, hence G = F. This proves that every element of G is conjugate to some element of D. Let H ' be some other trunk of G. In H' take an element h' which generates a dense subgroupof H ' . Then h' = 6h with suitable h E b,a E G ; obviously even h E D. Now 6 - ' H ' is a trunk in thecentralizer of h. Since the 1-component of the centralizer of h E D coincides with H , it follows that 6 H ' = H , which proves the remainder of Theorem 3 1.5. 31 .I 0 It is overdone to use dimension theory because F* is a rather elementary set to which elementary methods apply. This is done by the following Device Let P be a finite polytope of dimension p in real s-space S, and Q a closed subset in a q-dimensional linear subspace R' of real r-space R, with p + q < r. Let f be a continuous mapping of P into R. Then arbitrarily close tofthere is a continuousf* such thatf*P does not meet Q;f*may even be chosen to agree withfon a given closed subset Poof P such thatfP, n Q is void. Proof fPo and Q have a positive distance a. Choose a finite subpolytope PI

(of a subdivision of P), which does not meet Po and which contains theJ original of the +a-neighborhood of Q . Choose another finite subpolytope P2 of P, which does not meet Po either and which contains PI in its interior. Then f(P2\P,) has a distance > f a from Q. Choose a continuous real function u such that u(x) = O for x EP\P,, u(x) = 1 for x € P I , and 0 G ~ ( xG) 1 for x E P,\P,. A simplicia1 mapping f o of P2 into R is defined such that f o x =fx in the vertices x ofP2\P, and such that for any set of m 1 vertices ao, . .., a, (m~ p ) of PI the plane throughfoal (i = 0, . .., m) does not meet R' (which is q-dimensional). Thenfop, will not meet Q. With a fine enough subdivision of P and withfox nearfxin the vertices, one can make sure thatfo is an E-approximation off for a given E > 0. Finally one definesf*x = .fox + (1 - u ) f * x for x E P2andf*x =fx for x EP\P,. Then f* is still an E-approximation ofJ Furthermore,f*P, = f o p , does not meet Q , nor doesf*(P,\P,), as soon as E < +a.

+

31.

THE CONJUGA('Y THEOREM

I59

31.11. Proposition Let Q, (i - I ,2, . . .) be compact subsets of Cun-pieces of codimension =; 3 i n G, andf'a continuous mapping of ;I finite polytope T of dimension :: 2 into G. Then there is ; i n arbitr;irily close continuous approximationf* off'such thatf'Tdoes not meet Ui Q i ;/ * may even be required to agree withf'on a closed subset of TwhoseJiimiige does not meet Ui Q,.

Proof This statement is proved first for one summand, denoted by Q. It is allowable to interpret Q and a suitable open neighborhood B of it as lying in real r-space R , with r = dim G, where B is an open ball, and Q is a subset of a linear subspace R' of R with codimension 2 3. Take a smaller open ball B' that still contains Q. Form a finite subpolytope P of T such that j P is contained in B and P contains the )original of B'. Then, according to 31.10,,f restricted to P is approximated by an .f*such thatf* P does not meet Q, and agrees with Jon thef-original of B\B,' and finally f * is extended to the remainder of T, where the same agreement is required. This proves the statement for the case of one summand. A metric is assumed in G. Let E > 0 be given. Suppose a (1 - 2-")~-approximation f , off has been constructed and a closed neighborhood Sihas been assigned to Q, (i = 1, . .., n) such that f,T does not meet U;=, Si.Then 6 > 0 is chosen smaller than 2-"-' E and smaller than the distance between f T a,id U;,l S,; according to the first part of the proof, a 6-approximationf,,, off, i q constructed such thatf,+l Tdoes not meet Qn+l.Finally, a closed neighborhood S,,,, of Q,,, is chosen such that S,,,, is not met by fT. f * = limf;, fulfills the main requirements of the statement for a given E > 0. The supplementary one is easily satisfied. Remark If U, Q iis closed, the induction actually stops after a finite number of steps. This phenomenon can easily be accounted for in the proof.

31.12 What matters in the proof of 3 I .9 is the knowledge that GF \* is connected. This is now derived from 3 1.1 1. If x , y E G\F*, the connectedness of G guarantees the existence of a continuous mapping of the unit interval T into G with f ( 0 ) = x, f(1) = y. According to 3 I . I I , applied to F* (see 3 1,8),f is changed into a continuous mapping f * of T with f * ( O ) = x , f * ( I ) = y such that f T does not meet F*. This shows that GF\* is connected. Proposition Let p E F, U open in G. Then a path in G starting at p and finishing in U is p-U-homotopic with one in F. Two such paths contained in F, ifp-U-homotopic in G, are alsop-U-homotopic in F. Actually, as to the first part it has been proved only that in an E-neighborhood of a given M',, E W p uthere is a it', E W p usuch that ~ ' ~ (E7F)for all T. This, however, is easily turned into a homotopy, for example by putting

160

28-38.

TOPOLOGICAL AND INTEGRATIONMETHODS W J T ) = (exp

ua,)wo(T),

where aT,continuously depending on T , has been chosen such that exp a, = w , ( T )

W~(T)-~,

which is possible for small E 0. To prove the second part the former argument must be applied to a unit square rather than a unit interval. This proposition is used in the next section. 31.13. Historical Note The methods and results in this section are essentially H. Weyl’s in Math. 2 . 2 4 (1926) = Selecta, 348-352.

32. THE FUNDAMENTAL GROUP OF CENTERFREE U N ITAR I LY R ESTRICTED SEM IS1MPLE G ROUPS

G E Gru Lie Lin S S , centerfree (in 32.1-6), unitarily restricted. Other notations as in Section 31. It is shown that the fundamental group is finite. The fundamental group is computed in all relevant cases, although a simpler method will become available in Section 46. 32.1 As the set of left cosets aH of H in G, G / H bears a natural topology (see 4.8.1): the UH with U open in G are the open sets of GIH; by 12.12 GIH inherits a Can-manifoldstructure from G, compatible with this topology. The q E GIH map H by

qh=@

if g e q .

It is obvious that this definition is unambiguous. Proposition TG/H,Dl is a wrapping of F by means of y r4,hiqh. Actually it is the universal wrapping; that is, GIH is simply connected. This is proved in Proposition 32.1 1.

Since GIH is compact, 28.17.3 merits no attention (see 28.18), and it suffices to verify 28.17.2. According to 31.6-7,

Proof

Y rB.h 1 gh is locally topological at ‘I,hol for ho E D as a mapping of rEo,Dl into G (it maps a neighborhood of ‘1,ho1 topologically onto one of ho). YsgH being topological at 1 as a mapping of Eointo GIH, the preceding sentence may be restated as follows:

32.

THE FUNDAMENTAL G R O U P

161

32.1.I y rq.hiqh ( q E G / H , h E D ) is locally topological at rH,hol for ho E D. The proposition states that even

32.1.2 y r q , h i q /(~4 E G / H , h E D ) is locally topological at rqO,hol for qo E GIH, ho E D. This assertion follows from the fact that by left multiplication G acts transitively as a group of homeomorphisms on G / H : lfqo = g o H , then left multiplication by got in the first component, followed by y r4.h-iqhand by conjugation by means of go in G, sends 'qo,hol in this succession into rH,/?o', /I,, and 9ohog,' =qOho. All three mappings are at least locally topological i n corresponding points; their composite is y rU.hiqh.

32.2 In the metric induced on G by -$J (the opposite of the Killing form) every open ball is invariant under G, since the Killing form is itself invariant under G'. Now take in this metric 32.2.1 a smooth ball U around 0 (see Definition 8.4). I n particular, U is invariant under G', exp maps U homomorphically onto exp U = U , and TU c U for any real T with 171 < 1. U is simply connected, and so is U. A slraight path on U is meant as the exp-image of a rectilinear motion in U. With the notation of 32. I , q( D n U ) c U for q E G / H . I n the sequel p is a fixed point in D.

32.2.2. Proposition

it' E

W,,"is p-U-homotopic with a path on D.

By 31.12 it may be assumed that w(7) E F. After extending w, if needed, one may suppose that w( 1) E D.By 32.1 the equation

Proof

4 4 = d1 .

h(T)

can be solved by a path q on G / H and a pnth h on D withq(0) = H a n d h(0) = ~ ( 0 ) p . NOWput ,C,(T)

=q(O)h(T).

By 28.1 1 IV

is homotopic withq(0)h o qh(1);

/?(I) is a conjugate of ~ ( 1 ) by w(l)==q(l)h(l). By 32.2.1 it follows from II'( I ) E U that / I ( I ) E U , and for the same reason q ( ~ ) h ( lE) U . Consequently, 4/?(1) is / I ( I)-U-homotopic with a constant path. Furthermore q(0)k = h, since q(0) = H.

162

28-38.

TOPOLOGICAL AND INTEGRATIONMETHODS

This shows that w is p-U-homotopic with h,

which is situated on D as required by the proposition. 32.2.3. Proposition If wo,w1 E Wp",lying on D,are p-U-homotopic in

G , they are so in D . Proof

By 31.12 thep-U-homotopy of w0,w1 is already realized within F: wu(7) E

F,

w1(d E D,

WO(T),

wu(0) = p ,

w,(l)

E

u.

By 32.1 and 28.19 the equation M'u(4 = S U ( T > h

U ( 4

may be solved by q,,(T) E G / H , h , ( ~ )E D , depending continuously on with q,(O) = H , h,(O) = w,(O) = p. The solution is unique, hence hl

ho = M'O,

r~,~l,

= w1.

Finally h,( I) E U as a conjugate of w,( 1). So Y, h, meets the requirements for the p-U-homotopy of wo,w1 within D. 32.2.4. Definition 2 is the set of h distinct points 0 = zo,zI,. .., z k .

E

b with exp h = 1 ; 2 consists of the

Let Cibe a component of D n I !which accumulates at 1 (actually all of them do if U is small enough, but this does not matter). C, = log C, makes sense (see 31.4).It accumulates at some element of Z and every element of Z occurs in this quality once and only once. If, moreover, U is small enough, C, has only one accumulation point in Z . (Actually U is small enough.) So the C, may be numbered such that z, is the only accumulation point of C, in 2. Now let a path w €'?TIbe given. After a homotopic change (if necessary), one may assume that w = w I o w2 with w,(1) = p. Furthermore w I may be supposed to be straight in b,from 1 to p . By p-CI-homotopy w 2 is changed into w , lying on D (see 32.2.2) by means of w, (2 Q u Q 3). Straight paths u, (2 G u G 3) within U, with u,(O) = wu(l),u,(l) = 1, are attached to w,. Then w1 o w, o u, (2 G u G 3) describes a homotopical change of given w into w' E%'-~ such that

'

w'(0) = ~ ' ( 1 )= 1,

w'(.r) E

D

for 0 < T < 1 ;

w' is defined by ~ ' ( 7=)

log ~ ' ( 7 ) for 0 < T < 1,

w'(0) = 0,

w'(1)

= lim T=

1

log ~ ' ( 7 ) .

32.

THE FUNDAMENTAL G R O U P

163

~~'isap:ithotib,startitig~itOatide~iditig at soniez, E Z . It is homotopic o n b to a straight path from 0 to z,, called 12,. Then II,', and consequently I,', is hotnotopic with exp h,. So the exp h, represent the homotopy classes. It remains to show that they represent cvery homotopy class once. Suppose that exphi, exp hj are homotopic in G. Let u be a straight path i n 6, starting at logp and finishing a t 0. Put vi = u '1 hi and v j - u h,. Then exp ) p i . exp v j are p-I-homotopic and therefore p-U-homotopic; p i , Y , can be slightly changed into w i , w j , such that w i ( 0 ) = w j ( 0 ) = log p , and both are completely contained i n D . As vi( 1) = I z i ( l ) = t i , v , ( l ) = h j ( l ) = z j , still w i ( l ) E C i , w , ( l ) E C,. Now exp w i , exp w j are p U-homotopic i n G. By 32.2.3 they are already so within D. The endpoints of exp w i , exp w j are in Ci,Cj. During the homotopic transition they stay in U and in D. Therefore they stay in a component of D n U. Consequently Ci = C, whence i =j , which was asserted. The results are stated in the following theorem: I

Theorem Let G be a centerfree unitarily restricted semisimple group. The fundamental group of G is finite; its universal wrapping (and any wrapping whatsoever) is compact. Let D be a principd domain of C,and 2 the set of h E fi with exp h = 1. For any zi E 2 let hi be the straight path from 0 to zi. The exp hi form a representative system of the %'; '-homotopy classes. Sitice a unitary restriction o f a complex semisimple linear Lie group wraps a unitary restriction of its adjoint, it follows that the unitary restriction of any complex semisimple linear Lie group is compact.

32.3 To know the fundamental group

(lj of G explicitly more information is needed. Using hj rather than exp h, to indicate an element of @, one is advised to describe the group operation as an addition rather than a multiplication. By 29.1 it is known that exph, o exph, is homotopic with (exph,)(exph,) = exp(h, + h,). One must compute hi h,. Unfortunately, it need not belong to 6. This difficulty can be overcome by the use of the reflections S , (see 20.10). In 33.1 it will be shown that S , is induced by an inner automorphism of G . Inner automorphisms do not change the homotopy class of a path 11' E w,',since it' and c", M' are homotopic by 'f, c",wif yoc, is a path from 1 to c,. To study the action of such an automorphism on H , a basis a,, . . ., a, of H is chosen such that pr(aj)= O for k # j ,

+

=i

for k = j ,

where p,, .. ., p L are the elements of W++.

164

28-38.

TOPOLOGICAL AND INTEGRATION METHODS

describes the action of S,, in H*;

Put

Then (S,,, - 1)a,

=

Z a p . k a,, for k =j, P

for k # j.

=O

Let v, be the straight path from 0 to 27ra,. (The paths [I/ with i # 0 form a subset of the set of v,.) Then exp v, E W l l .Since exp v, and exp S,,,v, are homotopic, one recognizes that exp u,, with 'J

=

Z

vR,

P

is homotopic with the constant path. For Ccom E A, this produces the homotopically trivial paths exp u,, with u1 = -2v1

+ v2, . .., u, =

vj-1

- 2v,

+

v,+1,

..., u, =

v1-1

- 2v,.

A simple computation shows: exp v, is homotopic with exp j v l , exp(l

+ l)vl is homotopically trivial.

For CcomE D, one finds the homotopically trivial U I = -2v1

+ v,,

u j = Yj-1

- 2v,

u2 = -2v2

+

Vj+I,.

+

VI,

u3 = -2v3

.., u, = Y1 + P2 + V I - 1

+ Y4, .. .,

- 2v,,

which shows that exp 2 v l , exp 2v2, expo - 2)v3, exp(v,

+ v2 + v,)

are homotopically equivalent. 32.4 The fundamental group of a Cartesian product equals the direct product of those of the factors. To find the fundamental group of G, one may restrict oneself to the case of simple G. Then by 25.6 there is a top rootform which surpasses or equals all others in all coordinates. This shows:

32.

THE FUNDAMENTAL GROUP

165

Proposition If G is simple and p is its top rootform, a principal domain is

given by Impj>O

(j=l,

..., I ) ,

Imp~27r;

D is bounded in H by hyperplanes and Zconsists of all points h with pj(h) = 0 mod 274

Im p,(h)

2 0,

Im p(h) G 27r,

that is, apart from 0, one point for each primitive rootform occurring with multiplicity 1 in the top rootform. Using the top rootforms in Table E, one gets the following list for Z, in which the elements of Z are indicated by their pj-values as coordinates, omitting a factor 2ni 32.5 A,:

‘0,O,..., 0,01 .. ., 0,Ol ‘0,1, .. .)0,Ol FI,O,

‘O,O, B,:

c,:

...) l , O ’

‘O,O, ..., 0 , l l ‘O,O, ..., 0,01 ‘0, I , . . .)0,Ol ‘O,O, . . ., 0,O’ ‘O,O, ..., 0,l1

D,:

‘O,O,O, ‘l,O,O, ‘0,1,0,

..., 0,01 ..., 0,Ol

..., 0,Ol ‘0,0,1, ..., 0,Ol E,: ‘O,O, O,O, 0,01 ‘1,0,0,0,0,0~ ‘0, 0, I , o,o, 01 ‘O,O, 0, 0, 0, 0,O’ E7: ‘0, 1,0, 0, 0, 0 , O l E,: rO,O,O,O,O,O,O,O1 F,: ‘O,0,O,O1 G,: ‘0,Ol.

32.6 For A, the computation in 32.3 shows that the fundamental group is ( I 1)-cyclic. For D, the only nontrivial elements are exp Y,, exp v,, exp v3, all of order 2 if I is even, whereas for odd I, exp v, is of order 4. This shows :

+

Theorem Let G be a centerfree unitarily restricted simple semisimple Lie group. It is simply connected if GcomE G,, F,, E,. The fundamental group of C is two-cyclic for GcomE B,, Cl,E7, three-cyclic for GcomE E,, and ( I + 1)cyclic for GcomE A,. It is four-cyclic for GcomE D, ( I odd) and noncyclic of order 4 for GcomE D, (1 even).

32.7. Theorem Let G be a unitarily restricted semisimple linear Lie group. The center of G is contained in any trunk of G.

e

e

Proof G and G are isomorphic, and G and are locally isomorphic. As is centerfree, G wraps C by means of a homomorphism A. If H ( H ) is a trunk of G (C), AH ( A H ) is a trunk of Let w be a path in C leading from I to some center element. Then Aw is a path in G with (Aw)(O) = (Aw)(l) = 1. By 32.2 it is homotopic with a path

(e).

166

28-38.

TOPOLOGICAL AND INTEGRATIONMETHODS

Y Texp $ ( T ) withh(7) E A. Here y,h ( ~may ) be assumed to be continuous. Now , is mapped by h into y, exp @ T ) , is homotopic by 28.19, y, exp h ( ~ )which with w in G. It finishes at iv(1) which is contained in H . This proves the assertion. 32.8. An immediate consequence: Proposition Let G be a unitarily restricted semisimple linear Lie group. Then under the adjoint mapping the original of a t r u n k is again a trunk.

32.9. Another consequence: Conjugacy Theorem 3 1.5 remains valid for arbitrary unitarily restricted semisimple linear Lie groups.

32.10. Proposition The one-dimensional connected locally closed subgroups of a unitarily restricted semisimple linear Lie group G cover G. Indeed, a trunk of G is covered in this way, and by the conjugacy theorem this property is extended to G itself. 32.11. Proposition G/H is simply connected. Remark This result was announced in 32.1. It suffices to prove it for centerfree G, since the center is anyhow factored out according to 32.7.

Proof Let U be as in 32.2.1, and Ci as in 32.2.4. Since U is c-invariant, rG/H, D n Ul is the original of U n F under the wrapping of F by r G / H , D1. The r G / H , Cil are the different components of r G / H , D n U l . Since U\Fhas codimension 3 in U, U n F is still simply connected. Hence each rG/H, Cil wraps U n F homeomorphically. Consequently the number of points above a point of Fequals the number of the Ci, which is the cardinality of the fundamental group. So rG/H, D1, and G/Has well, is simply connected. 32.12. Historical Note The preceding approach to the fundamental group is H. Weyl’s, Math. Z . 24 (1926)=Sefecta 348-352, though the actual computation of the fundamental group was performed by E. Cartan, Ann. Mat. 4,209-256 (1927) = Euvres I, 2,793-840. 33. THE AUTOMORPHISMS OF SEMISIMPLE LIE GROUPS

G is assumed to be semisimple in ordered third dressing, complex or unitarily restricted, and sometimes centerfree or simply connected. Other notations as usual.

33.

167

THE AUTOMORPHlSMS

33.1. Reflections as Automorphisms

A (continuous) automorphism of G leaving the trunk H invariant, induces an automorphism of W * . Conversely, any automorphism of W * is induced by one of G, which extends to a local one of G , to a global one of the universal wrapping, and (by factoring out the center) to a global one of the adjoint group. A special automorphism of W * is known as S, ( a E W * ; see 20.10). Proposition S , is induced by an iiiner automorphism of C (unitary or complex), which interchanges e, and e-,. Proof

For a

E

W * put

Then Pa

4 2 ( a , a>)"2qa,

qa

par,

= +(2(cr,

Pa h = 0 With

T =

a))112 r,,

h

forall

E

H

with

a(h) = O .

77i(2(a, a))-*, TPa q a = -niqz,

+,r,

= mr,,

7 j j ,h = O

for h E H

with

a(h)=O;

hence ( ~ XTP P a) qa

=

-qa

9

(exp 7 P a ) ra = -ra, (exp TP,) h = h for h Since 2h,

= qa

E

H with a(h) = 0.

+ r,, (exp T P , ) ~ ,= -ha.

Therefore exp TP, maps

which mapping in fact induces S,. It also interchanges e, and e-,.

28-38.

168

TOPOLOGICAL AND INTEGRATION METHODS

exp TP, is an inner automorphism of complex G.Because of third dressing, it is even an inner automorphism of unitarily restricted C, since T is purely imaginary. Definition exp(ni(2(a,a))-* (2, noted by S,.

+ E,)), as an extension of S,, is also de-

33.2-3. The Automorphisms of W* 33.2.1. Definition Aut W* is the group of automorphisms of W*; also if linearly extended to H * or H: and if transferred to H a n d HsI.The subgroup generated by the reflections S, (aE W*), considered as mappings of W*, or H *, H,: or H , HsI, is called the kaleidoscope group or group of inner automorphisms of W*, indicated by Int W*. The group of automorphisms of thegraph Wtt is denoted by Aut W++;also if linearly extended to H * or H; or transferred to Hand HsI. As a permutation group of W*, Aut W* is finite. Int W* is normal in Aut W*. Aut W* preserves the inner product in H.: 33.2.2. Proposition Int W* is generated by the reflections S, with PE

w++.

Proof By25.4 every a E Wt can be obtained as a Tpo,where po E Wt+ and T is a product of reflections S,, with p E W++.Obviously S, = TS,,T-', which

proves the statement. 33.2.3. Proposition Under Int W* every element of W* is equivalent to one of W++. This follows as well from 25.4. 33.2.4. Proposition If C is simple, then Int W* acts transitively in each of the (at most two) sets of nonzero rootforms with equal lengths. Proof By 33.2.3 it suffices to show that primitive rootforms of the same

length are Int W*-equivalent. Inspection of the graphs shows that the subgraphs of primitive rootforms with the same length are connected. Now, if p , u are adjacent primitive rootforms with the same length, then S,,S,,p = u. This proves the proposition. 33.3 Every A E Aut Wt+ somehow extends to complex G,say

Ae,

= v,

eA,.

33.

THE AUTOMORPHISMS

169

Using an /I E H with p(h) = log v, for all p E W + +and arbitrary log-values, one can replace A by Ah-' with h = exp h, in this way reducing all v, ( p E W + + ) to 1.

33.3.1. Definition An extension of A Ae,

for all p

= eAp

E

E

Aut W f + to G is called straight if

W++.

Proposition A E Aut W++possesses a straight extension to G.Because of 25.4.2, it is unique (after ordered third dressing). It commutes with C,,, Cst,M . It leaves the unitary restriction invariant. The straight extensions of the elements of Aut W++to G clearly form a group.

33.3.2. Definition The group consisting of the straight extensions of Aut W + +to G is denoted by AutC(Wt+). It is a subgroup of Aut G. An automorphism of G need not be induced by one of G.

33.3.3. Definition The subgroup of Aut G consisting of the automorphisms inducing elements of AutC( W++)is denoted by AutC(W++). If G is centerfree or simply connected, any automorphism of G is induced by one of G and therefore all of AutG(W + + )is induced by AutC( W++).Therefore they are isomorphic. 33.3.4 Aut W++interchanges the components of W++.For simple G,that is, connected W++,it is trivial in most cases. The exceptions are A, (l > 1) : inverting the graph, D , ( l > 4): permutingp,,p2,

D,:

permutingp1,p,,p,,

E,:

simultaneously permutingp, and p3, p4and ps.

Comparing these data with those on the fundamental group (32.5), one notes that nontrivial elements of-Aut W + +induce nontrivial automorphisms of the fundamental group of centerfree Gun.On the other hand, Int Gunalways acts trivially on the fundamental group. So nontrivial elements of Aut W++ cannot be induced by elements of Int Gun, whereas all Int W* is induced by elements of Int Gun(see 33.1). This proves: Proposition Int W * n Aut W++= (1).

In 33.8 this is proved in a more elementary way by avoiding the fundamental group.

170

28-38.

TOPOLOGICAL AND INTEGRATIONMETHODS

33.&5. Chambers 33.4. Definition The open convex domains in which H: is divided by the orthoplanes a'- of the rootforms a # 0 are called chambers. Evidently Aut W* permutes the chambers. Any natural basis Z (see 25.2) defines a chamber C(2): 33.4.1

5 E C(2)e,(5, p) > 0 for all

p E Z.

Indeed, any a combines from Z with all coefficients nonnegative or nonpositive; therefore (5, a)= 0 for no a E W* as long as .$ E C ( Z ) , and thus no plane aL enters C ( Z ) . Definition 5 E H: is called dominant if, in the given order, it is surpassed by no other Int W*-equivalent element of H:. A dominant chamber is one with a dominant element. If .$ is in some chamber, then any maximal element of the set (Int W*) 4 is dominant and again in a chamber. Therefore there is at least one dominant chamber.

33.4.2

(5,p) 2 0 for all p

E

W++

is necessary to make 5 dominant, for, if ( 5 , p ) < 0, then S, 5 = 5 -2((5,p)/(p,p))p is higher than 5. The system of inequalities 33.4.2 describes the closure of C( W"), as defined by 33.4.1. Therefore every dominant element is in C ( W + + ) and every dominant element that is in some chamber is in C(W++).Consequently there is only one dominant chamber, namely, C( W++). If C , is some chamber, then for 5 E C , any dominant equivalent of 5 is contained in (Int W*)E n C(W++). Therefore (Int W*)C, n C(W++) is nonvoid, as is C, n (Int W*)C( W"). This shows that in every chamber (Int W*)C(W++)has some point, and, since Int W* permutes the chambers, the following obtains : 33.4.3 Int W* is transitive on the set of chambers.

Now consider an element of Aut W* that preserves the dominant chamber C. It interchanges the walls, and thus the elements of W++,while preserving

the angles, and so it preserves the structure of the graph W++. It belongs to Aut W++. By the proposition of 33.3.4 an element of Int W* preserving C must be 1. Hence : 33.4.4 Int W* is simply transitive on the set of chambers.

33.

THE AUTOMOKPHISMS

171

33.4.5 A consequence: Different points of a chamber can never be Int W*-equivalent. In 33.8.1 this is extended to closed chambers. Another consequence: No element of Int W : k ,except the identity, can have a fixed point in a chamber. So if S E Int W* is a reflection, its reflecting plane cannot cross any chamber. Hence: Proposition Any reflection belonging to Int W* has the form S , with some a €

w*.

33.4.7 The foregoing results are expressed i n the following theorem: Theorem I n t W* is generated by the S,(p E W ' ')and is simply transitive on the set of ch:imbers. A complete representative system of Aut W*/lnt W* is furnished by Aut W' '. Any element of Int W* is induced by some inner automorphism of C leaving H and C,, invariant, but no element # I of Aut W ' ' is induced by an inner automorphism of G. Any element of Aut W* is induced by some automorphism of C leaving H a n d G,, invariant. The last assertion is easily verified: if A E Aut W*, then A = A , A, with A , E Aut W + ' ,A , E Tnt W*, where the extensions of A , , A , are achieved as in 33.3.1 and 33. I . 33.5. Proposition If Z is a natural basis of W * , the intersection of the halfspaces ( 5 , ~>) 0 with p E Z is a chamber. By this construction natural bases and chambers are one-to-one related. Under a suitable order of H$ any chamber may play the role of dominant chamber. The structure of the graph of Cdoes not depend on the choice of the order in H:.

ts

The first assertion was proved in 33.4.1. Int W* preserves the notions of natural basis and chamber. It is transitive on the set of chambers. Therefore every chamber is produced by a natural basis. It is evident that this relation is one-to-one. According to 25.2.5, any natural basis can be obtained as a W t + under a suitable order. In this way any chamber can appear as the dominant one. All chambers are congruent, and thus all possible graphs are isomorphic. Proof

The last sentence of the proposition fills the gap indicated in 25.3 and 26.24.

Remark

33.6-8. An Alternative Approach 33.6. Proposition If a,P E W* and 2 .r(a,p) = 2x/nz,,p, then S,Sp is a rotation of order m,.p around the intersection of m i and pl as the axis. In particular, S,Sp = Sp S, if and only if a = i p or ( a ,p) = 0. I f cz # ip,the only possible values of m,,p are 2,3,4, and 6 .

172

28-38.

TOPOLOGICAL AND INTEGRATIONMETHODS

The proposition is obvious. Actually, the relations sp*= 1, ( SP sO)mp.o = 1 (p # u), between the generators S, (T E W++) of Int W* characterize Int W* as an abstract group. (Without proof.)

33.7. Proposition Let B be the intersection of the two half-spaces

t&t,p) > 0, t&, u) > 0 (p,u E W + + ,p # 0 ) and let Kp,Obe the group generated by Sp,S,. The interior angle of B is 7r/mp,,(see 33.6), and the Kp,,-images of B do not meet each other.

Proof The first assertion follows from the fact that 3: (p, u) = 7r( 1 - (mP,,,)-'), X ( p , u ) being obtuse, and the second from the first and Proposition 33.6 because this proposition implies that the orbits of B and SPBunder SpSueach consist of mp,Oelements, all mutually disjoint.

33.8 In the proof of the proposition of 33.3.4, on which part of Theorem 33.4.7 rests, the fundamental group has been used. A much more elementary proof is possible. Eventually one has to show that different points of the dominant chamber Ccannot be Int W*-equivalent. Even more will beproved:

c cannot be Int W*-equivalent. In other words, 33.4.2 is also sufficient for 5 to be dominant.

33.8.1. Proposition Different points of 33.8.2. Proposition If

t E C, then

surpasses all its other Int W*-

equivalents.

33.8.3 Derivation of 33.8.1 from 33.8.2. Suppose 33.8.2 and both t and S t in C(some S E Int W*); then, by 33.8.2,t 2 Sf as well as S t 2 S - ' S t , which proves 33.8.1. 33.8.4 If 1 , E~H: are Int W*-equivalent, they can be joined by a chain

t=to,tl,..., f P = ~suchthat foreveryj=l, ...,~ : ~ , # ~ , - ~ , a n d ~ , = S ~ ~ , - ~ and 6, agree. Only the for some p E W + + .In all but one W++-coordinate coordinate corresponding to p may decrease or increase from to t,. It will be shown: Proposition If toE C,then increases can be avoided.

(It is evident that this proposition implies 33.8.2.) Proof For the needs of the proof assume a total order on H:, extending the

given one (see 25.2.7). Since ( 5 , ~> ) 0 for all p E W++, no S,,t is higher than 5; thus t, is definitely lower than to.If the chain to, fl,..., 5, shows some increase, then there must

33. THE AUTOMORPHISMS

173

be a reversal-Ej, which means a Ej such that Tj-, > 6, < Ej+,. Suppose that the chain between 5 and 7 has been chosen such that the lowest reversal-f, is as high as possible. This will lead to a contradiction. Let 6, be the lowest reversal, Ej

=

spS,-,,

Ej+i

(Ej,

SuE j ,

1

P) < 0,

with p, 0 E W++, < 0.

( E j , 0)

Put $0 = E j i

$,+, = S,,&

for even i,

$,+, = SUBi

for odd

i,

up to 9 2 m p . o = $0,

with mp.udefined as in 33.7. The intersection of ft(E,p) > 0 and t5(E,a) > 0 is called B as in 33.7. In the $-sequence a reversal from decrease to increase could take place only at ~9~E -B. Since the K,,,-images of B do not overlap (see 33.7), no 9, E - B except 6,. Thus there is no such reversal in the $-sequence. to the other way The &sequence is now modified by traveling from around the intersection of p* and ul, that is, through =a,,a,, ..., QZmp.o-l = f , + , . The lowest reversal-[, has been removed, while no lower reversals have been created. The lowest reversal has been raised, which proves the proposition by contradiction.

33.8.5 A consequence of 33.2.4 and 33.8.1 is the following: Proposition If C is simple, then among the rootforms of the same (nonvanishing) length there is only one dominant one. This is also confirmed by Table E.

33.9. Automorphisms of C and G Definition

Aut(C, H ) Aut(G, H ) Int(C, H ) Int(G, H )

is the group ofautomorphisms of Cleaving Hinvariant, is the group ofautomorphisms of G leaving Hinvariant, is the group ofinner automorphisms of Cleaving Hinvariant, is the group of inner automorphisms of G leaving Hinvariant;

A is used for both the subgroup of Aut G as well as that of Aut G constituted by the exp h with h E H.

28-38.

174

TOPOLOGICAL A N D INTEGRATION METHODS

Theorem For complex or unitarily restricted G:

(1) Int G is transitive on the set of trunks of G, (2) fl equals the group of automorphisms of G, leaving H elementwise invariant, (3) Aut(G, H ) / A 2( Aut W* in a natural way, (4) Int(G, H ) / f l 21 Int W* by restriction of the preceding mapping, (5) Aut(G, H) = AutC( W + + .Int(G, ) H), (6) Aut G = AutC(W++) .Int G, with trivial intersections of the factors. Furthermore the following statements hold :

(i’) (i = 1,2,4,5,6), obtained by replacing G,H by G,H in the statement i, and (3’) Aut(G, H ) / Ais isomorphic to a subgroup of Aut W*, which in the case of centerfree or simply connected G again coincides with Aut W*. Proof (I), (1’). See the conjugacy theorems 17.8, 31.5, 32.9.

(2), (2’). It suffices to consider an automorphism f of G with fa = u for all E Hsuch that f = exp k. Clearly, for some v,

u E H and to find an h

fe,

= v,

e,.

From the commutator relations it follows that (a)

v, v-,

= 1,

V,

(a, p, a

vfi = va+fi

+ p E W*).

ForhEH he,

(exp k)e,

=

a(h)e,,

= (exp a(h)) e,.

To prove the assertion one must determine h E H such that (b)

exp a(h) = v,.

There is an h E H that fulfills (b) for all a E W + + .Then by (a) it fulfills (b) for all a E W*,since the e, (a E W + + )generate G (see 25.4.2). In the unitary case one must take care that h E iH,,, but in that case f respects the unitary restriction, which means v,e,

=fe, = C,,

fCu,e,

=

-Gun fe-,

=

-Cu,v-,

= -P-,

hence v,=v-,=v,

--I

.

e-,

C,, e-,

= P-,

e,,

33. THE AUTOMORPHISMS

175

Now, since Iv,I = 1, all a(h) are purely imaginary for h fulfilling (b). So iH,, = Hun. (3)(4)(3')(4'). f E Aut(G, H ) or E Aut(G,H) induces an element Of of Aut W*. This inducement 0 is a homomorphism. Its kernel consists of the f that induce the identity on W*, hence on H. By (2) this is just H. Conversely, by 33.4.7 any element of Aut W* (Int W*) is induced by some of Aut(G, H) (Int(G, H)) (whether G is complex or unitarily restricted), and this is again induced by some of Aut(G, H) (Int(G, H)) if G is centerfree or simply connected (see 33.1). (5)(5'). The image of Aut(G,H) or Aut(G,H) under the inducement O contains Int W* and is contained in Aut W*. By 33.4.7 the splitting Aut W* =Aut W++-IntW* is lifted to Aut(G,H), where, according to 33.3.2-3, the first factor can be assumed to be in AutCW++. By this procedure Of can be lifted into f'=fl f2 with some fl E AutC(W"), fi E Int(G,H) or fi E AutC(W++), f2 E Int(G,H). Now Of' = Of; hence by (2) f ETA. Therefore f E fl Int(G,H), respectively, Efl Int(G,H), which proves the assertions. (6)(6'). Given f E Aut G or E Aut G, there is, according to (I), some g E Int G, respectively, Int G, which carries H (H) into the trunk fH (fH) By (5)(5') g-'f E AutC(W++).IntG, respectively, AutG(W++).IntG, which proves the statement about$ The statement on the intersection of the factors is obvious. h

E

33.10. Automorphisms and the Fundamental Group Proposition An automorphism of centerfree G is inner if, and only if it acts trivially on the fundamental group of G (the unitary restriction). Proof The automorphism f may be ,assumed to leave the trunk H invariant and even to preserve the dominant chamber, thus to induce an element of Aut W++. If it is outer, then 33.3.4 states that it acts nontrivially on the fundamental group. If it is inner, it acts trivially according to 32.3. 33.11. Plus-Automorphism In the case of simple G (see 33.3.4), with the exception of D4, there is at most one nontrivial element of Aut W + + ,which extends to dressed G as a straight automorphism according to 33.3. For later use it is convenient to have the following: Definition The plus-automorphism P is determined after ordered third dressing as the straight extension of P defined on H * by A, (,> 1): Dl ( I > 4 ) : €6:

ppl

PP, = PL+I-,, PPI = P 2 9 p P 2 = P l , p P , = P , ( j # = p-33 pp3 = PI, pp4= PS, ppS = p4, pp2

1 9 %

= p23

pp6 = p6.

176

28-38.

TOPOLOGICAL AND INTEGRATION METHODS

P is involutory; it commutes with the minus-automorphism M , with C,, and C,,,and, except for D,, is uniquely determined. 33.12-1 3. Minus-Automorphism 33.12. Proposition For simple G the minus-automorphism M is inner, except in the cases A, (I > l), D, (I odd), E,, where it is outer. M preserves the simple summands of semisimple G. Proof M maps an element of the fundamental group of centerfree unitary G

into its inverse. By 33.10 and 32.6 the assertion is obvious. 33.13 As mentioned in 23.1, the given third dressing is not uniquely determined by its trunk H . Assume that M ' is an involutory automorphism of G acting on the trunk H as the scalar multiplication by -1. Then by Theoiem 33.9

M'

=

h, M for some h , E H ,

which can also be written as M'

= &&-I

=

M,

where h is some (existent) square root of h, in H . Putting ei = Ae, (a E w*), one gets

M'ei

=

h M k l .he, = he-,

#(he,, he-,)

= +(ea, e-,)

= ei,, =

1.

In other words, a new third dressing on His obtained with M' as its minusautomorphism ; its unitary semimorphism is hC,,h-', since hC,,h-'ek = hC,,h-' .he, = -he-, = -e'-,, and its unitary restriction is &Gun. More generally, let M ' be an involutory automorphism of C acting on some trunk H' of G as the scalar multiplication by -1. Since trunks are conjugate, there is an a E G such that H' = a"H. Then a"-' M'a" is as in the first situation, and a"-'M'a"= &MA-' for some h E H . In the same way as above it is shown that M' is the minus-automorphism of a third dressing on H' with branches Jhe, (a E W*), unitary semimorphism &C,, h-'a"-', and unitary restriction a"hG,,. It follows:

33.

THE AUTOMORPHISMS

177

Theorem The unitary restrictions and unitary semimorphisms of G with respect to the various third dressings form a conjugacy class under Int G and so do the minus-automorphisms. The possible minus-automorphisms are the involutory automorphisms of G that act as scalar multiplications by -1 on some trunk .

A n involutory automorphism and a trunk, as mentioned in the theorem, determine a unitary restriction, but an involutory automorphism alone does not. Different involutory automorphisms with the same trunk may yield the same unitary restriction. Remark

Proposition A unitary semimorphism of Cis a unitary semimorphism with respect to any trunk left invariant by it. Proof

I t may be assumed that the unitary semimorphism is the given one C,, defined with respect to H . Let H ' be another trunk, C , , H ' = H ' . By 32.9 there is an element a E G,, such that H ' = ClH. Then, if M is the minusautomorphism of the given third dressing, the involutory automorphism ClMCl-' defines a third dressing on H ' with unitary semimorphism ClC,,a"-' =

33.14. The Corner Lattice 33.14.1 The notion of principal domain (3 1.4) can be generalized: Definition The trunk H,, is divided into domains called fundamental, by means of the hyperplanes defined by CY = 0 mod 2n-i ( a E W*). The fundamental domains are bounded and convex, and one of them is the principal domain. The notion of fundamental domain is carried over to H,, by exp. 33.14.2 The chambers are one-to-one related to the fundamental domains with a corner 0 (namely, by multiplication by i and inclusion); Int W* is simply transitive on the set of chambers (33.4.4). 33.14.3. Definition The corner lattice consists of the points /J E H,, with ~ ( h=) 0 mod 27ri for all cc E W* (it suffices to require it for a E W"), 33.14.4. Definition The group generated by Tnt W* (acting on Hun) and the translations of the corner lattice is denoted by Int mod W*.

28-38.

178

TOPOLOGICAL AND INTEGRATIONMETHODS

33.14.5. Definition SU*,is the reflection in the hyperplane a = m.2ni (m integral, CL E W*).The group generated by the &,, is denoted by Int' mod W*. 33.14.6 Int mod W* interchanges the fundamental domains. It contains the &,,; thus Int' mod W* sub Int mod W*. Proposition Int' mod W* acts transitively on the set of fundamental

domains. Proof Let D' be a fundamental domain, h E D', and w a path from 0 to h in Hun.After a slight modification, it may be assumed that if w leaves a fundamental domain it will pass through the interior of a face and into an adjacent domain. Let the faces successively crossed by w lie in the hyperplanes ak = mk.2qi (uk E W*, where k = 1, . .., q).Then D" = S,,,,, S,,,,, D' is a fundamental domain with a corner 0. By 33.14.2 there is an S E Int W* such that SD" = D . This proves the assertion.

Remark One can show that Int' mod W* is even simply transitive on the set of fundamental domains.

33.14.7 The parallelepiped Q c Hundefined by

h E Q t+ 0 < Im p(h) < 27r

for all p E W++

is mapped one-to-one into If,, by exp; its closure is mapped onto Hun. The division of Q into fundamental domains is transferred to If,,,,. By exp an element of Int mod W* is mapped into one induced by an element of Gun, since this is true for the elements of Int W* and since the translations in Int mod W* are mapped trivially. Hence: Proposition Hunis divided into sets equivalent to b under two of them meet in boundary points only.

G,,,such that

33.16. Kaleidoscope Groups

A,, B,, Cl, D,, The terminology on H , as introduced in Section 16 (see also Table C), is used. A,: Sprw, SpPI

=w, =Wltlr

S p , w l t l = 9.

for j # i , i + l ,

34.

INTEGRATION IN COMPACT GROUPS

179

The reflections generate the symmetric group of / + 1 permutands. B, S p I w i= w l

s,,wi

for i # / ,

-w1,

as Sf,-, i n

S,,, (i # I )

The general element of Int W* is a permutation of wl, . .., w , combined with a mapping w 1 -+ f w , , . . ., w , -+ +w,, the signs being independent of each other. C , : S,,

( i f I)

S,,wi = w i

as in A,, for i # I,

s,,w , = -w1* The same Int W" as i n B,. as S,,., in A,-,, Sp,wi=Sf2wl=wi for i < l - 2 ,

D,: S,,

(i# 1,2)

s,,wi-i = - w , ,

Sp2w1= - W , - l , S,,w1-1

=w,,

s,,w , =

WI-1.

The general element of Int W * is a permutation of w I , . . ., w I combined with a mapping w I + * w , , . . ., wi + ko,, with independent signs such that the number of minus signs is even. The case of the exceptional groups is more involved. Without proof: the orders of Int W * are G,: 12, F,: 1152, E,: 72.6!, E,: 5 6 * 7 2 * 6 ! , E,: 2 4 0 3 6 . 7 2 . 6 ! 33.17. Historical Note Though implicitly used by E. Cartan, the explicit introduction of the kaleidoscope group is due to H. Weyl, Math. Z. 24 (1926) = Selecta, 338. The kaleidoscope groups of the exceptional groups were more closely studied by E. Cartan, Amer. J . Math. 18, 1-61 (1896); Ei/n.es I 1,293-353. Another method is expounded in: H. Freudenthal, Proc. Kori. Akad. Wet. A57,487-491 (1954) = Indagatiories Math. 16.

34. INTEGRATION IN COMPACT GROUPS 34.1-2. Measure in Groups 34.1 In a real linear Lie group G an infinitesimal measure (see 5.4) assigned to the tangent space at 1 can be transferred by multiplication on the right

180

28-38.

TOPOLOGICAL A N D INTEGRATIONMETHODS

to any point of G. Such an infinitesimal measure is invariant under multiplication on the right, as is the measure arising from it by integration. It is important to know whether this measure is invariant under left multiplication as well. This is the case if and only if Z (a E G ) preserves the infinitesimal measure at 1 or, equivalently, if and only if det a" = 1 for all a E G. If G is semisimple, this condition is fulfilled, for then tr a' = 0 for u E C . It is also fulfilled if c" is compact because then det maps c" onto a bounded connected subgroup of the multiplication group of reals, which can consist of 1 only. 34.2 Without proof it may be mentioned that a one-sided invariant measure exists in any locally compact group (Hum measure); if the group is compact, it is two-sided invariant and the total measure is finite. A short proof is given of the existence of such a measure in compact groups G, or rather of the existence of an invariant average of continuous functions on G.

34.3. Measure in Compact Groups

Let G be a compact group, and @ the linear space of real continuous functions of G topologized by means of the norm I.. .I defined by

Linear mappings 9,, Ya(a E G ) of @ onto @ are defined by . ,' and y a y # are homomorphisms of G. Note that af'% W,f,respectively 6pGf, is the set of 9,f9 ,, f (a E G). The closure of the convex envelope of BGf,ZG f is denoted by CBcf, C Z G J Because of the uniform continuity of continuous f,

C9,A Cdp,f are compact.

Furthermore, they are convex and invariant under all

W,,respectively 9,; osc = max - min

is a continuous functional on @. If there exists anything like a continuous invariant average on 0, it has to be the same for all B,f, hence for all elements of C9,f. If a constant function can be found in CBcf, its value should be the average off. To find such a constant one looks for the function with the smallest oscillation and proves that its oscillation vanishes.

34.

INTEGRATION IN COMPACT GROUPS

181

Put inr o s c g = y . !7ECXG.f

Because of the compactness of C.OA,f, there is a g

E

C.9c;f with

osc g = y . Given some E > 0, because of the compactness of C%',g, there are a,, ..., a,,,

E

G

such that for every a E G there is a j with Jg(sa)- g(xa,)l < E

Putting

for all

.Y E

G.

c

h = ( 1/m> *POL g, one obtains osc h < osc g. On the other hand,

It E

C.P,fi hence osc h > osc g.

Thus osc h = osc g.

Further. min g G min h < max h < max g; hence max h = max g. Now take .yo such that h(so)= max h. Then g(soak)= max g

For a

E

for

k

=

1,

. . ., m.

G there is a j with

I d . U o a) - d . Y o a,) I

E,

and t h LIs g(s,a)

rnax g - s

for every

a E G.

This is true for every E > 0. Therefore g is constant. The same reasoning applies to CPG.fi The value of a constant function in C.4'G CYc;f.is called a right, respectively, a hfi, werqsy of/-: Let A , B be right and left averages ofJ Then

28-38.

182

TOPOLOGICAL AND INTEGRATION METHODS

Ix u,f(xa,) - '4 I < - BI < 3E, Ix for a suitable choice of aj,bk G and real uI, ..., u,, PI, ..., P,, such that a, 2 0, 2 0, a, I , x 1. Replacing x in the first equality by b;'x, multiplying by and summing up, one obtains Ix aJf(b;l xu,) - I < jE, 3E,

PkB,f(b;l

E

=

Pk

Pk =

Pk,

Pk

and likewise

Ic

Pk

aJf(b;l xu,) - BI < $E*

From this it follows that every right average offequals every left average off, and therefore all are equal. Similarly, it can be shown that the average A is a continuous linear functional on @. Obviously A is invariant under gRand -YRbecause C g Gf and C-YGf are invariant under gRand -YR,respectively. Furthermore, iff(x) > 0 for all x, then Af>0. Even more can be proved : iff(x) 2 0 for all x andf(x) # 0 for some x, then Af>0. Indeed, in this case there is an open U c G on which f(x) > 6 for some 6 >O; G is covered by a finite number of Uu,say Uul, . ,., Uu,. (gRk-, f)(x) > 6 20

for x

E

uuk

elsewhere;

hence, if g = ( l/m) 2 gRk-, f,

then ..ug = Af,

and g(x) > (l/m) 6

for all x

E

G.

Finally, A g = A ( g - ( l/m) 6)

+ A ( l/m) 6

( l/m) 6 > 0.

Theorem On a compact group G there exists an average d ( o r AG) of continuous functions with the following properties. 34.3.1

A?is linear.

34.

183

INTEGRATION IN COMPACT GROUPS

34.3.2

A9a=AYa=A for a E G ,

34.3.3

A1 = 1,

34.3.4 If f ( x ) 2 0 for all x

E

G and f ( x ) # 0 for some x

E

G, then d f > 0.

JY is continuous.

34.3.5

The properties 34.3.1, 34.3.2, 34.3.3, 34.3.5 characterize A. So do 34.3.1, 34.3.2,34.3.3,34.3.4,even if the last one is weakened to f ( x ) 2 0 + Af> 0.

34.3.4'

The last remark is easily proved by observing that 34.3.4' can be generalized by 34.3.1 and 34.3.3 to -E

= Af= A Y b A

Y a f(a, b>.

The average A extends in an obvious way to complex functions and finitemaps the finitedimensional linear-space-valued functions. Then, if dimensional linear spaces R into R' linearly, 34.3.7

Mo?f1= qJAf

for linear functions from G to R . The above characterization shows that if p is an invariant measure on G, gauged by p(G) = 1, then for continuous f

conversely such an invariant measure can easily be recovered from the average. 34.4. Historical Note The preceding construction of the average in compact groups is due to J. von Neumann (Compositio Math. 1, 106-114 (1934)). Haar's measure dates from 1933 (Ann. Math. 34, 147-169). Simpler constructions are extant (see, e.g., S . Banach in S . Saks, Theory of the Integral, 1937; A. Weil, L'intdgration duns les groupes topologiques et ses applications, Hermann, Paris, 1940, 1951).

184

28-38.

TOPOLOGICAL AND INTEGRATIONMETHODS

35. THE CONDUClBlLlTY THEOREM 35.1-2. An Invariant Inner Product 35.1. Theorem Suppose R E Spa Lin Rea or Com, dim R < w, G sub Aut R. If G is compact, then R can be endowed with a definite hermitean inner product, invariant under G (which then becomes a group of unitary mappings). Proof Starting with some definite hermitean inner product (. .., define a* for a E End R by

.. .) in

R,

for all x , y E R ; a* depends continuously on a, and so does a* a. s = d& yaEGa*a makes sense as an element of End R. Furthermore, s* = s, (ax, y ) = (x, a*y)

(sx, x )

Y ac&* ax, x) = -4Y a s G ( a X , ax),

=4 3

which vanishes only if (ax, ax) = 0 for all a, hence if (x, x) = 0. Therefore = (sx, y ) may be used as a new definite hermitean inner product on R. Moreover, (cx, cy)’ = (SCX, cy) = (c*scx, y ) , and, using the invariance property of A? for c E G, s = d G gCyaEG a* a = d G YaeG(aC)* (ac) = A?G c* (ya& a* a) c = c* sc, which proves the invariance of the new inner product. (x, y)’

35.2. Proposition Suppose R E Spa Lin Rea or Corn Inp, G sub Aut R. Then for every complete G-invariant linear subspace S of R there is a G-invariant linear subspace T of R, such that R = S + T is a direct sum, namely the orthoplement of S in R. The proof is obvious. Note that the completeness of S implies that its orthoplement is a linear complement in R even if R is not complete. 35.3-4. Conducibility 35.3. Definition Suppose R E Spa Lin, 0 =A Y

ada)rn

= (%

$1.

36.8. Proposition Up to equivalence every continuous irreducible linear representation f of compact G in R E Spa Lin Com (dim R < w) can also be isolated from 9by restricting the latter to an invariant linear subspace of @. Proof Choose a linear functional u # 0 on R and define Qf,,, as the set of Yauf(a)x. Then @f,u sub @. By

k = YxY.uf(a)x R is mapped onto Of,,,.Now k is linear, and # O because u # 0. kx = Y a u f ( 4 x,

?gC kx = yaU ~ ( U Cx) = ' f a u ~ ( u ) ~ x( c=) kf(c) X ; thus

.Bck = kf(c). The kernel of k is invariant underf(c) for all c, hence = (0).Thus k provides an restricted to @f,u). equivalence offand y, (.gC Definition For any continuous finite-dimensional linear representation f o f G call @f,u the set of Y.uf(a)x; y,(9= restricted to @f.u) is calledf,. The linear space spanned by Gf ,", if u ranges through all linear functionals on R, is called Qf; or, alternatively, Qf is the linear space spanned by the matrix coefficients off on some basis of R.

190

28-38.

TOPOLOGICAL AND INTEGRATION METHODS

For equivalent f, @, is the same; 65, splits in correspondence with the inequivalent irreducible components off: 36.9 I n 36.8 it was shown that any finite-dimensional continuous irreducible linear representationfof compact G can be recovered within @, namely, asf;, acting on @f,u. Now it is shown: Proposition Letfbe a finite-dimensional continuous linear representation of compact G in R. Then any S sub @ acted on by 9 equivalently to f is some @ f , u (on which W acts a s h ) . Proof

Let k be the given equivalence; thus &?a

k

=

kf(a) for all a E G.

Put u=YXER((~+W)).

Then

( k x ) ( ~=) (gakx)(l)= (kf(a)x) ( I )

=

uf(a)x.

Thus

k x = y, uf(a)x, k R = 65fvu.

36.10-1 1. Orthonormality 36.10 By 36.9 the linear subspaces of @ which are acted on by 9 2 equivalently to irreduciblef are contained in 65, (see Definition 36.8). Thus as a consequence of 36.6: Proposition Tffand g are inequivalent, then @, is orthogonal to

Qg.

36.11. Proposition R E Spa Lin Com Inp, dim R = n ; an orthonormal basis of R is given; f i s a continuous unitary irreducible representation of compact G in R . Then the matrix coefficients offas elements of @ are mutually orthogonal, each showing the norm n-lI2.

36.

191

ORTHOGONALITYRELATIONS

defines a sesquilinear form on R, which is invariant underf(a) for all a Now (x, y)’ = (hx, y ) for some h E End R . (Ax,

v) = (hf(4x, f(4Y ) = (f(4-’

E

G.

4Y ) ,

h =f(a)-1 hf(a). By 36.3, h turns out to be a scalar multiplication, Therefore the ratio (x, Y)’l(X? Y )

does not depend on x , ~ though , it may still depend on u,u. By a symmetric argument:

Y (f(4XY 4(f(4Y , u) = Y ( X 9 Y )(u, V)Y

36.11 .I

II

where y is a constant. A coordinate function is a Y x ( x , u ) , where u is a basis member. A matrix coefficient off has the form Yx(f(a)x,u), where x, u are basis members. For different pairs rx,ul,‘y,u1 of basis members 36.1 1.1 vanishes, which proves the orthogonality of different matrix coefficients as members of @. Summing up 36.1 1.1 over x = y, u = u ranging over a basis, one gets, after the d#l? Y,-sign, the sum of absolute squares of the matrix coefficients of a unitary matrix, which is n. The second member becomes yn2, which proves y = n-*. Substituting this into 36.1 1 . 1 one confirms the assertion on the norm.

36.12-1 5. Totality 36.12. Proposition For compact G the Qf,with f running through a complete set of inequivalent continuous irreducible finite-dimensional linear representations of G, span a linear subspace @’ of @ which is dense in @, even in the uniform topology. The proof (see 36.14) uses standard methods of the theory of integral equations, expounded in 36.13. 36.13 Let R

E

Spa Lin Com Inp, with inner product (..., ...) and norm

11.. .I1 defined by llxll = (x,x)”2.

For a linear mapping A of R into R,

ll4= as in 2.2; A is boundedif llAll < a.

SUP IIAxll,

IlXU 0 the eigenvectors of L belonging to eigenvalues h with

Ihl > y span a finite-dimensional space R,. Proof If this were not true, there would be an infinite orthonormal sequence of xj with Lxj = Xjxj, lhjl 2 y. Since L is relatively compact, one may suppose

that the sequence of Lxj converges. But ~ILXj sk,(l

= iihjxj - h k x k l l = ( h j 2

+

hk2)"2

> y1/2

f o r j # k , which is a contradiction. 36.13.6 By 36.13.5 any orthonormal set of eigenvectors belonging to nonz 2 , . . ., be a maximal one. Then vanishing eigenvalues of L is countable. Let zI, for x E R m

Lx

=

c (Lx,z,) z,. 1

The restriction ofL to the orthoplement Ry' of R , is still hermitean and relatively compact. It follows from 36.13.3 that IlLuIl < yI1uII for u E Ry'. Put x, = x - C'/(x,z,)z,. Now, for almost all n, x, E Ry' (which follows from 36.13.4-5), whence liLx,ll< y/lxnllG yllxll. This is true for any y and almost all n. Thus lim Lx, = 0. Now Proof

n

n

LX,

= Lx

-

c ( x ,z,) Lz, 1

=

Lx -

c (Lx,z,) I

z,,

=

Lx 1

( x ,Lz,) z ,

194

28-38.

TOPOLOGICAL AND INTEGRATION METHODS

which shows m

Lx =

(Lx,z,) 2,. 1

36.13.7. Lemma Let [...I be another norm on R, with llxll Q 1x1 for all x E R, and assume that the image under L of the set of x with llxll Q 1 is even relatively compact in the norm I**.I. Then, with x E R and z I , z 2 ,..., as in

36.13.6, OD

Lx =

with respect to

(Lx,z,)z, I

[*.*I.

This follows from: 36.13.8. Lemma If I...) is as in 36.13.7 and lim a,, = a in 11-. 11, sequence of a, is relatively compact in I * * I, then lim a,, = a in I * -1.

-

and the

Proof Every subsequence of the a,, has a convergent subsequence in I...[ whose limit cannot be anything else but a. If the a, did not converge to a

(everything in the sense of 1- -.I), there would be a subsequence that would not have a as a limit point, hence a subsubsequence converging to a point different from a, contradictory to the first observation. 36.14. Proof of 36.12 Lemma 36.13.7 is applied to the space @ of continuous functions on compact G (instead of R ) , with

(v, 4) = ”4 Y d a ) a3 as an inner product and 1 ~ =1 max, lv(a)l as the I-..l-norm; @is I-.-l-complete. II

A continuous function 6 on G is chosen with 6(u-1)=6(a)

for u E G

and Ls (instead of L) is defined by L s v = Y.”4Ybwb-1)v(b)= YII(v,%16). Obviously Lb is hermitean (see 34.3.6). Because of the uniform continuity of 6, there is for any given E > 0 a l-neighborhood U in G such that (9tC6 -6 1 < E for c E U . Now for llvll Q 1

6 - 9 I I 2 - 1 611 Ilvll-l%7,-16 Q p ? a 2 1 1 1 - l 6 - 9.1 < E

I(L8v)(al) - (Lbv’)(az)l= 1(%%1-1

- .%2-1

61

as soon as a2 E Ual. This shows that the set of functions Lav with (1vl1Q 1 is equicontinuous and I...l-bounded, hence relatively I--.l-compact, since @ is I.-.l-complete. Therefore Ls is relatively I-..l-compact.

(9&*L$9TCcp)(U) = ”4 y b 7 q U c - l b-’)q(bc) = AY*6(ab-’)p7(b)

= (Lgv)(a),

37.

THE CHARACTERS OF COMPACT GROUPS

195

hence Ls, Wc= WcLa.

Therefore, if Lgq = AT,

then Ls, gCq= AWcq.

Thus the A-eigenspace of Ls, is WG-invariant. The restriction of W to the A-eigenspace of Ls, is called W A(A # 0). This is a continuous finite-dimensional unitary linear representation of G. By 36.9 the , , , contained in @@A and therefore A-eigenspace of Ls, (A # 0) is some Q ~ A , thus belonging to continuous in @', which was the linear space spanned by all finite-dimensional linear representations f of G. By the lemma La@ is contained in the I..*l-closure of the linear space spanned by the A-eigenspaces (A # 0). Thus Ls,@ C @. Now for every q E @ there is a 19 such that ILgcp - ql is arbitrarily small: make 8(a) > 0 for all a E G, a(a) = 0 outside a small 1-neighborhood in G, and ,A8= 1. Consequently, Us,Ls,@= @ and therefore @= 0,all closures being taken in the (***l-sense. This proves Proposition 36.12. 36.15 The preceding results are summarized in a theorem: Theorem For compact G, in the sense of the invariant average, the matrix coefficients of a complete set of inequivalent continuous irreducible finitedimensional unitary representations, if divided by a factor n-'I2 (n = dimension of the representation), form an orthonormal system, which is total in that every continuous function on G can be uniformly approximated by finite linear combinations of the functions of the system. 37. THE CHARACTERS OF COMPACT GROUPS 37.1. Orthonormality Definition The trace of a finite-dimensional linear representation of a group is called a character of G. It is the same for equivalent representations. A character is called continuous, irreducible, or n-dimensional, depending on whether a representation from which it can be derived has this property.

The continuous characters of compact G belong to @ (see 36.7). Definition The elements of @ that are constant on every inner class of G are called class functions. The continuous characters are special class functions.

196

28-38.

TOPOLOGICAL AND INTEGRATION METHODS

Theorem The continuous irreducible characters of compact G are ortho-

gonal if the respective representations are not equivalent, and all among them have unit norm. They form a total orthonormal system for the subspace of class functions in the sense that every class function can be uniformly approximated by finite linear combinations of these characters. Proof The first assertion is an immediate consequence of the fact that a character is a sum of the diagonal matrix coefficients with respect to a basis. Let y be a class function. As an element of 0, up to a given E > 0, it equals a finite sum a j k f Ya(f(a)

X I ' ,

Xk'),

where f means a continuous irreducible linear representation of G in some R, with the basis xlf, ..., xi,. Now

which commutes with all f(b) (b E G), is a scalar multiplication according to Proposition 36.3. The multiplier, found by tracing, is

n;' tr f ( a ) . Now, since y is a class function, y(a) = "4 Y c v(c-' ac),

which up to E can be replaced by

4 Y c z Q,k,(f(c-'

ac>x,f, Xkf) =

c

a/k/ n;'

t r f ( 4 (x,',

Xk3.

This proves the second assertion. 3 7 . 2 4 Reduction t o the Trunk

37.2 G is now supposed to be a centerfree unitarily restricted semisimple linear Lie group. The notions and notations are those of Section 31 ; G is known to be compact and therefore 37.1 may be applied. The average A on G can be calculated by integrating with a suitably gauged infinitesimal measure dP.

The conjugacy theorem (31.5) states that a class function is determined by its behavior in the principal domain D . One may ask how the orthogonality properties of characters, as formulated in 37.1, can be expressed in terms of the restrictions of the characters to D or to H.

37.

THE CHARACTERS OF COMPACT GROUPS

197

Theorem Every invariant infinitesimal measure dp on G is related to an

invariant infinitesimal measure dv on H such that for any continuous class function 0

lG 1 @(h)J-J

(1 - exp a(h))dv(h),

@ dp =

UEW*

h and h being related by exp h = h.

Proof When integrating a function 0over G, the complement of F(see 31.9), which is nowhere dense, does not matter; F i s wrapped in rG/H, Dl by means of y r a , h , qh (see 32.1). This wrapping transplants 0 to rG/H, Dl . Under the condition that the infinitesimal measure dp on G is also transplanted from F to rG/H, Dl as an infinitesimal measure dj2 the integration can be performed in rG/H, Dl just as well; of course, the number of sheets s has to be taken into account. Note that if 0 is a class function its transplant 8 to rG/H, Dl in a point ‘q,h7 depends only on h. Another than the transplant measure dj2 is better adapted to integration in. rG/H, Dl: Near H one can identify I:with TG/H,D l by means of 37.2.1

gh + ‘4, h’

with q = g H .

Then, heuristically, rG/H, Dl touches G tangentially along D (identified with r H / H , Dl). Now by 37.2.1 dp is transferred from Galong Dto rG/H,Dl along rH/H, Dl . The new infinitesimal measure dp* is invariant under H applied to the second member within the r,l-brackets. By left multiplication on G/Hit is carried over all rG/H, Dl . dp* = dp, -dp2is a Cartesian product measnre in rG/H,Dl with dpl as a Ginvariant infinitesimal measure on G / H and dp2 as an H-invariant one on H, at present restricted to D. Along r H / H , D l it coincides with that in F by the identification 37.2.1. If 0 is a class function, then

which reduces the integration to one in D.How to calculate dj2/dp* at some point of r H / H , D l becomes clear from the diagram

-

wrapping rq,hlqh q=gH t II q = g H gh

gh

‘p

28-38.

198

TOPOLOGICAL AND INTEGRATION METHODS

The transition by the vertical arrow is that of 37.2.1 which preserves the infinitesimal measure. Therefore their behavior under wrapping is described by the mapping called rp in 31.6. This gives

dci (h) = det grad,, rp = dP*

(1 - exp a@))

if h = exp h E D.

UEW*

This shows

where y does not depend on 0.(Note that exp a(h) depends on h rather than on h.) By 33.14.7 an inner automorphism of G leaving H invariant changes b in a fundamental domain to which it is equivalent under Int mod W * while preserving the differential measure and the value of the integral. Summing up over all equivalent fundamental domains, one gets, essentially, the integral over H ; hence

where y’ can be made 1 by gauging dp2 suitably. Because of its H-invariance, dp2 is an invariant infinitesimal measure on the I-dimensional torus H. In the sequel it is indicated by dv.

37.3. Definition Q is defined as a function on H(or rather on H ) , by

and for any continuous irreducible finite-dimensional linear representation

f of G, =

Q(h) tr f(h) if h E H , h = exp h.

Since a@) is purely imaginary, one gets

-

Q(h)=

TI

USW-

(1 - ~ X a@ P)).

From 37.1 it follows: Theorem With a suitably gauged invariant infinitesimal measure dv on H

0 for inequivalent 1 for equivalent

Therefore the irreducible characters multiplied by Q form an orthonormal system with respect to the ordinary measure on H (total with respect to class functions). An algebraic proof for this result is given in Section 48.

37. THE CHARACTERS OF COMPACT GROUPS

199

37.4 The results of 37.3 remain valid for an arbitrary unitarily restricted semisimple G (not necessarily centerfree) : G is a wrapping of c' by means of a certain A. The measures dp and dp in them are related to one another by invariance up to a constant factor under the wrapping h as far as h is one-to-one. A class function 0 on G defines a class function 0 on c' by means of

= const'

&)

n(1- exp a(h))di@),

where h = exp h. Because of 32.7-8 and a(h) = a(@, this may be replaced by

which proves the assertion. 37.5.t The Natural Gauge The tangent space G of compact semisimple G is naturally gifted with a positive inner product by the negative of the Killing form. In turn, this inner product in G fixes a natural elementary volume, namely, by assigning the value 1 to an orthonormal basis. This leads to a natural invariant infinitesimal measure and to a measure on G, independent of the choice of the basis. One might wonder how large G,, measures with this natural yardstick. 37.5.1. Definition The natural measure po on compact semisimple G is determined by an invariant natural infinitesimal measure dpo that assigns to some (-$,,)-orthonormal basis the elementary volume 1. 37.5.2. Problem To determine the total po-measure-value of G. It suffices to solve it for centerfree simple G. In H an elementary volume v', which assigns the value I to a (-#IH)-orthonormal basis, is introduced. It determines an invariant infinitesimal measure dv in H . With a slight deviation in terminology from Theorem 37.2 the invariant measure p in C is determined such that for every class function 0

IG

0dp =

t

ID@(A)

(1 - exp a(h))du(h). aEW*

The contents of 37.5 are not used in the sequel.

200

28-38.

TOPOLOGICAL AND INTEGRATIONMETHODS

Anticipating a formula to be verified in 47.5.3,one gets p(G) = k v ( D ) = kv'(D),

where k is the order of the kaleidoscope group. v'(D) can easily be calculated. Let G be given in ordered second dressing, p l , . . .)pI the primitive rootforms, and 2 q r p fthe top rootform. As in 32.3,let a', . . ., a, in H be determined by

Then D is the I-simplex with vertices 0 and the 27rq;' aj( j = 1, . . ., I). Its volume according to v is

1 ~

I!

(24'(det r- #(aI,a J ) 1 f . j = l ) 1II / 2 4;' j

1

=-

I!

(2n)'(detr(pi,pj)1f,j=1)-1/2nq;'. j

This leads to the formula : 37.5.3

k

P(G) = ~ ( 2 ~ ) ' ( d e t r ( p , , p i ) ' f , J = 1 ) - 1l-rqj'. '2 j

It is clear how this formula has to be modified for nonsimple G. pO(G)is computed by comparing po with p. Put

37.5.4

B, means a ball around 0 in G with radius E , according to the (-#un)-metricy and BE is its intersection with H . The characteristic function of exp B ; is a class function. For small E

where the dots represent a term of order €'+I, and r = dim G .

37.

THE CHARACTERS OF COMPACT CROUPS

20 1

On the other hand, a well-known formula assesses the volume of B, as

hence

This gives

In order to be evaluated, the integral in the denominator is replaced by 37.5.6

Corresponding to the r-I=2m

elements cc of W”, independent real variables (, are assumed and A ( [ )=

SH c+‘”JyC E, a(h))2“dv‘(h) OL

is considered. The coefficient of

4, in A ( [ ) , written as a polynomial, is just (2m) ! 1.

To compute A ( ( ) , orthonormal coordinates ulr . . ., u1are assumed in Hsuch that uI is a fixed multiple of ,$,a; hence

c

UI =

(C5, a , c 5, m)-1/2

i

c f , a.

Then

c =(-I>”(c c 5,cc)”’.2l-

A ( ( ) = ( - I P ( f~a x ,

( , a ) ‘ ~ S ~ e - c J l l J ( h ) ~ U : m d‘ U . *~c i u l

(/:e-”2du)l-’

202

28-38.

The coefficient of 7

TOPOLOGICAL AND INTEGRATIONMETHODS

na in (-l)m (C

=

[a

a, C I, a)m is

4

C(Qi1YQ i J ( % Y

* * *

(ah-1’

where the sum runs over all permutations i of W*. This gives

To come back to y, put

FA =

1,. l-r 4 ) d W .

Then

FA = A’ Fl

Now by 37.5.6

= +rFI

r(+r) = F, r(+r + 1). *

is now substituted into the denominator of 37.5.5 by using the value of I obtained in 37.5.7: r(l12)r 22mm! nm y=-i-

=

7

Now by 37.5.3-4: Theorem With the natural measure po for centerfree simple G

37

THE CHARACTERS OF COMPACT GROUPS

203

where m = $(r - I), the qi are the coefficients of the top rootform on a natural basis p,, . . ., pl, and 7=

( e l l r nn a i v aiz)(%,

* * *

(%m-l,

%),

with the sum running over all permutations i of W*. Nonsimple G requires an evident modification. Unfortunately, the numerical value of T does not seem to be readily accessible, except for the simple groups of rank ~ 3In, fact,

1 = 2 : ~ = 2 ( m ! ) ~ (a,a); UEWt

hence

Furthermore, A,:

7

-

m ! aEW+

(a,a)=288.

0, :

= 6!*48.

c, :

=

6!.48.

37.6. Historical Note The Frobenius-1.Schur character theory for finite groups was extended to compact groups by F. Peter and H. Weyl, Math. Ann. 97 (1927) = Sefecta H. Weyl, 387-404. The formula for the natural volume of G was derived by H. Freudenthal.

204

28-38.

TOPOLOGICAL AND INTEGRATIONMETHODS

38. SOME GLOBAL PROPERTIES OF SEMISIMPLE LINEAR LIE GROUPS

G E Gru Lie Lin Com S S in third dressing, presented by linear mappings of E Spa Lin Com (dim R = n < m); therefore G c Aut R . Since by 32.2.4 Gunis compact, R can be provided with a Gun-invariantinner product (. .., ...). For a E End R one defines a* by (ax, y ) = (x, a*y) as before. R

38.1-4. The Hermitean-Unitary Split 38.1 The hermitean elements of End R are those a with a* = a, the antihermitean those with a* = -a; End R is the direct sum of the subspace S of the hermitean elements and the subspace U of the antihermitean elements, as follows from the involutory nature of *. The eigenvalues of an hermitean element are real, those of an antihermitean purely imaginary; both kinds ofelements appear in diagonal form on a suitable orthonormal basis of R . The hermitean elements of End R with positive (nonnegative) eigenvalues only are called positive definite (semidefinite); they are the a E S with ( a x , x ) > 0 ((ax,x) > 0) for x E R , x # O . Let S be the set of all positive definite hermitean elements of Aut R, U the group of unitary elements of Aut R, that is, of those a for which a* = a - l ; U is the infinitesimal algebra of U, and exp S c S. S is closed in Aut R, for if a is in the closure of S it is by continuity still positive semidefinite hermitean, but, since, a E Aut R, it has no vanishing eigenvalues and is even positive definite. If the kernel of the trace on End R is interpreted as a member of An-] (see 16.2), its intersection with U coincides with the unitary restriction formed according to 24.6. 38.2. Theorem exp maps S homeomorphically onto S, with a nowhere degenerate gradient, and S is a closed analytic submanifold of Aut R . This is a consequence of Theorem 18.1.4, and 38. I. 38.3. Theorem As an analytic manifold Aut R is the product of its closed submanifolds U and S by means of the multiplication in Aut R ; explicitly, every a E Aut R can be uniquely written as a = su, with s E S , u E U depending analytically on a. Proof

U is a closed submanifold because it is a compact Lie group; S is so

by 38.2. If a ~ A uRt and a = s l u l =s2u2 with st ES,uiE U ( i = 1,2), then ua*= s1uIul*sI*= s12 = s , ~ ;since division by 2 is unique in S, extraction of square roots is so in S ; hence s1 = s2 and u1 = u2.

38.

SOME GLOBAL PROPERTIES

205

If a E Aut R, then aa* E S ; now take s as the square root of aa* in S and u = s-la; then a = su as required, with s, and u as well, depending analytically on a. 38.4. Theorem As an analytic manifold G is the product of its closed submanifolds G,, and exp iC,, by means of the multiplication in G; exp iC,, = C n S, G,, = G n U , that is, exp iC,, consists of all positive definite hermitean elements of G, and Gunconsists of all unitary elements of G. Proof Because it is a compact Lie group, G,, is a closed submanifold, as is exp iC,, also because it is the image of iC,, = C n S under exp, Now G,;exp iC,, is a closed submanifold of Aut R , hence also a closed submanifold of G, but with the same dimension as G ; hence G = Gun* exp Gun. It also follows from the uniqueness in 38.3 that G n (I= G,,, G n S = exp Gun.

38.5. Fundamental Group, Center, Closedness Theorem

( I ) The fundamental groups of a complex semisimple linear Lie group G and its unitary restriction are isomorphic in a natural way. (2) The center of a complex semisimple linear Lie group G is finite. (3) The center of a real semisimple linear Lie Group G is finite. (4) A real or complex semisimple linear Lie group acting in R E Spa Lin Rea or Com is closed in Aut R . (Anticipated in 30.1 .) Proof

(I) G is homeomorphic with the topological product of Gunand exp iC,, in a natural way (38.4), and the fundamental group of exp iC,, is trivial. (2) The center may be interpreted as a factor group of the fundamental group of the adjoint G"; by the first part of this theorem and Theorem 32.2.4 the fundamental group of (? is finite. (3) The center of G is contained in that of Gc,,. (4) By Theorem 19.10 and Proposition 12.16 this follows from the finiteness of the center. Remark Do not draw the conclusion that every real semisimple linear Lie group has a finite fundamental group. The universal wrapping need not be realizable as a linear Lie group. The class of connected groups locally isomorphic to semisimple linear Lie groups is larger than that of those isomorphic

206

28-38.

TOPOLOGICAL AND INTEGRATION METHODS

to semisimple linear Lie groups. However, there is a universal linear wrapping for any G E Gru Lie Lin SS (see 62.9).

38.6. Historical Note The hermitean-unitary split is due to E. Cartan. It has still to be viewed in a broader context (see Section 64).

THE ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

39. THE ASSOCIATIVE ENVELOPE OF A LIE ALGEBRA

In the following general theorems on a Lie algebra G, the finiteness of the basis X is not involved, only its existence. The scalars need not come from a field; a commutative ring with 1-element is sufficient. The aim is to embed G into an associative algebra such that: 39.1

[a,b] = ab - ba

for a, b E G,

and to ensure that no other relations but 39.1 are introduced. 39.2 The basis X is supposed to be ordered (0. The free associative algebra with unit-element 1 generated by X is called d ( X ) ;G is identified in a natural way with a subset of d ( X ) . C is the subset of xy - yx - [x, y ] with x, y E X (or G). $(C)is the ideal generated by C in d ( X ) .

Definition &(G)= d ( X ) mod $(C)is called the associative envelope of G. (The dependence of &(G) on Xis inessential.) Clearly this is an associative algebra in which 39.1 prevails. It is not evident whether what corresponds to G in d ( X ) is mapped one-to-one into d(G)by the reduction mod $(C). However, a still stronger proposition will be proved. 39.3 A product of elements of X (monomial) is called orderly if its factors appear in the correct order; these terms will also be used forthescalarmultiples. An element of d ( X ) is called orderly if it is a sum of orderly monomials. Up to first degree monomials, y x can be replaced by xy mod $(C).Consequently, up to lower degree terms every monomial can be replaced by an orderly monomial mod $(C). By induction on the degree one can replace any element of d ( X ) with an orderly one mod $(C). Therefore every class of d ( X ) mod $(C)contains an orderly representative. It will be proved:

Theorem The only orderly element in $(C) is 0. In other words, the subspace of orderly elements of d ( X ) is a linear complement of $(C). Consequence As a subset of d ( G ) ,G is mapped one-to-one into b(G)by the reduction mod $(C). 207

208

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

39.4 Another property is the following. Theorem Let y be a mapping of G into any associative algebra A such that y[a,b]= p * y6 - ~ byu.. Then q~ can be uniquely extended to a homomorphism of B(G) into A. Furthermore, B(G) is characterized by this property.

39.5 A linear representation of &(G)will serve to prove Theorem 39.3. Let Xbe a copy of X.The element of Xcorresponding to one of Xis written as its light face counterpart. If x < y, then x < y, by definition. The free commutative associative algebra with unit-element 1 generated by X is denoted by Y ( X ) . The subset of elements of degree G n is denoted by Y n ( X ) .The set of monomials of degree n is denoted by .Mn(X). If x E Xand u E .Mn(X),then x G u means x G every factor involved in u. The next step is to define a multiplication

gc E Y ( X )

for g E G, c

E

Y(X),

in such a way that 39.5.1 y,gc is linear, if Y ( X ) is considered as a linear space. 39.5.2 (alg, + a 2 g 2 ) ~ = a I g l c + a 2 g 2 c f scalars or aI,a2. It suffices to define this product for g E X and c E U, A , ( X ) and to require that it be extended by the use of 39.5.1-2. The definition proceeds by induction on the degree. The basic assumption is

xl = x

for X E X .

Under the assumption that g u is defined for all g E G,u xu is defined for all x

39.5.3

xu=xu

39.5.4

xu = yxu

if

E

E

Yn-] (X),

X,u E .Mn(X)': XGU,

+ [x, y ] u

if not x G u,u = yu, and y G v. Note that 39.5.5

xw = xw mod Y , ( X )

It is shown by induction that

if w E Y n ( X ) .

39. 39.5.6

209

THE ASSOCIATIVE ENVELOPE

for x, y

xyu - yxu = [x, y ] u

E

G, u E Y ( X ) .

The induction step consists in assuming that 39.5.6 is true for u E Yn-l( A') and ascertaining it for u E ,nU,(X) while restricting x, y to X.One may add the assumption 39.5.7

y G x,

which means no restriction at all. 39.5.8 The induction step is allowed if, in addition, y G u. Indeed, then Q xu, hence

y

xyu

= xyu = yxu

+ [x, y ] u,

where the first equality sign is justified by 39.5.3 applied to yu instead of xu and the second by 39.5.4, with u instead of u. The induction step has still to be justified, if not y Q u. Then one can split 39.5.9

z E

u = zv,

x, u E .nU,-I(X),

39.5.10

z G u,

39.5.11

Z

39.5.12

z<x

a)(ha)*

Adding up these equations ( j = -1,O,

. .., i - I),

one finds

1

41.3.2

PI =

-C0 ( A +&)(ha)

= -(i = -(i

+ 1)(A + $ia)(ha) + 1)((A, a ) + +(a, a)),

because of second dressing (see 21.1.8). If the a-ladder x o , x l , ... is finite, say of lengthp, then pp = 0; thus, 41.3.3

In this case, 41.3.4

P I = 3(j+

1)(P - d ( a , 4.

As in 21.1.8, one can state the following.

216

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

41.3.5. Proposition A finite a-ladder of weights starting at h has length -2((h, a)/(a,a)). It is turned upside down into a (+)-ladder by S , (see 20.10, 21.1.8). All finite a-ladders differing by real multiples of a have the same midpoint. In particular, if h is a weight, a E W + ,and j , k are maximal and minimal integers such that h +ja, h + ka are weights, then S,(h +ja) = h + ka, and all h + p t with ~ k Q p ~j are weights. 41.4-5.

Integral Elements

41.4 The foregoing leads to a definition. Definition f

E H*

if -

is called integral, or

4 E Hi*,,,

2 is an~ integer for every a

( a ,4

E

w*.

Obviously W c Hi*,* c H;. The following has been proved. Proposition Weights in finite a-ladders ( a E W*) are integral elements of H:. 41.5. Proposition

5 E HEg iff -2((5,p)/(p,p)) are integers for all p E W++.

Proof “Only if” is obvious. The condition can be restated in this form: S, f - f is an integral multiple of p E W + + .If it is fulfilled, then clearly Sf - 4 is a linear combination from W++with integral coefficients for any S E Int W*. In particular,

is so for a = p,p, E W+.Thus all -2((f, a)/(a,a ) ) p , are integers, whereas by 25.4.1 the p j have no common divisor. Therefore -2((f, &)/(a,a)) is integral for every ct E W + ,which is the assertion of the proposition. 41.6-7.

The Weyl Tool

41.6. Proposition S, ( p W+\{P).

E

W + + )interchanges & p and is a permutation on

42. Proof

some u

SOURCE, TOP WEIGHT, AND LIMITATION

217

If ct E W+\(p}, then on the basis W + +a has a positive coefficient for W t t with u # p . So has

E

Hence all its nonzero coefficients are positive and SpaE W + . 41.7. Definition 6 = 4

zuEW+ a is called the Weyl tool.

Proposition 2((6,p)/(p,p)) = 1 for all p

E

W + + ;hence 6 is dominant and

6 EHig. Proof By applying Proposition 41.6 one gets Sp26 = 26 - 2p

for p

E

W++,

thus, 2-=1f ) (69

for P E W++.

(P 9 P)

42.

SOURCE, TOP WEIGHT, AND LIMITATION OF A REPRESENTATION

The notations are as before. 42.1. Source Definition If every G-invariant subspace of R containing x E R coincides with R , x is called a source of the representation of G. If x is a source and u , , . . ., a, form a basis of G, then R is linearly spanned by the aIlu/2* .a1,x.

-

Actually the ail

.:’

...

suffice, for every “wrong pair” ads, (c < d ) can be replaced by 12,414 + [ a d , ( I c ] , where the second summand gives rise to shorter products, On a basis of G using second-dressing branches, the 42.1.1

II

e-u I

. . , e‘rn

Urn ,]I OLI

. . .el,m,Hf~. . .)j?x

span R , where a,, . . ., a,,, are the positive rootforms arranged in some linear order, preferably compatible with the partial order in H:.

218

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

If, moreover, x belongs to a weight A, then in 42.1.1 the factors h may be canceled because of hi x = h(hi).x.Then the element in 42.1.1 (if # 0) belongs to the weight

and there are no other weights. This shows: Proposition If there is a source of weight A, then, for every weight p, the difference p - h is an integral linear combination from W*. If there is a source of integral weight, all weights are integral.

42.2-3. Source of Top Weight 42.2. Definition Given a representation, a weight exceeding the others, if it exists, is called the top weight. Suppose a maximal weight h and a source x belonging to h as its weight vector. Then e,x (ct E W + )vanishes, since otherwise it would belong to the weight h + a > A. Then all terms 42.1.1 with some j , > 0 can be left out of consideration. Therefore h is even the top weight; R is already spanned by the 42.2.1

e-p,

* *

e-pmx,

where p , , . , ., pm stem from W++in an arbitrary succession in which repetitions are allowed. This follows from the fact that every e-, ( a E W + )can be obtained from the e-p ( p E W++)as a scalar multiple of compound commutators. The weight of 42.2.1 (if # 0) is h - p4. It follows: 42.2.2. Proposition Suppose there is a maximal weight h and a source belonging to h. Then (1) h has the multiplicity 1.

(2) Every weight p has a finite multiplicity. (3) h is the top weight. 42.2.3 From 41.1.1 it is known what happens to weights under an automorphism T of G. If T leaves H and W++invariant, it preserves the partial order on H: and thus maps the top weight into the top weight. Hence: Proposition If h is the top weight of the linear representationfof G and T is an automorphism of G leaving H a n d W++invariant, then

7'-'h = y hsjy A( Th)

42.

219

SOURCE, TOP WEIGHT, AND LIMITATION

is the top weight of

Y aE G f(m.

42.3. Definition The representation with a top weight source vector x is limited if for every p E W++almost all e!,x vanish. Note that in case of limitation the (-p)-ladder ( p E W++)starting at x has lengthp computed by 41.3.3. Proposition

If the representation is limited, then for any y

E

R almost all

e i , y ( p E W + +vanish. ) For every p E W++every weight belongs to a p-ladder.

Proof One may suppose

-

y = e-p, * . e-pnx (pl, . . ., pm E W++as in 42.2.1) and then proceed by induction along m. The induction step is justified by the following statement:

If e l p y = 0 ,

then

el+P3e-, y = O

for p,u

E

W++.

This statement is true, since e::’ e-, y

+

e-p y Kel+’ P e-p-a y = = e - , e ~ + ,y~ ~ ‘ e - ~ - , e ! ! +y, ~ ~ “ e - ~ ~ - , , e ! ! +y, l = e!’p2 e-,

+

+

+

K

,,,e-3p-oe!!py = o

with scalars K , . . ., K ’ , K”,K ” ’ ; remember that -u - 4p cannot be a rootform. It follows that for every weight p and every p E W++there is an integer j > 0 such that p - j p starts a p-ladder. Let k be the maximal integer such that p + k p is a weight. Then Sp(p- j p ) G p kp, hence Sp(p kp) G p - j p s p < p + kp which shows that p belongs to the p-ladder starting at S p b + kp).

+

+

42.4. An Inequality Under the assumptions of Propositions 42.2.2 and 42.3 any p-ladder of weights (p E W++)is bounded from above according to 42.2.2 and from below by 42.3. Therefore it is finite. A still more efficient bound is exhibited in the following proposition. Proposition Let the representation be limited with top weight h and a source belonging to A. Then

+

+

( p 6, p 6 ) < (A for each other weight p (see 41.7).

+ 6, h + 6 )

For some p E W++let v = p + j p be the weight with maximal integralj if the weight p is given; 41.3.5 applied to Y instead of h and -p instead of a shows

Proof

39-50.

220

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

Using 41.7, one gets (v

+ 6, v + 6) - ( p + 6, p + 6) = (v + 6, v + 6) - (v + 6 - j p , v + 6 - j p ) =

-i2(P,

P)

+ 2XP, 4 + 2j(P, 6)

2 (-j2 + j 2 + j ) ( f , P )

As long a s p # A, there is some p (v

E

=j(P,p)-

W + +such t h a t j > 0; but thenj(p,p) > 0 and

+ 6, v + 8) > ( p + 6, p + 6)

for some higher weight v. This inductive argument does not stop until p = A. This proves the assertion. 42.5. Finite Dimensionality and Irreducibility If the representation is irreducible, every x E R,x # 0 is a source. If, moreover, it is finite-dimensional, the number of weights is finite, one is top weight, and the representation is limited. The converse is also true : Theorem Let the representation of G E Alg Lie Com SS in R E Spa Lin Com be limited with a top weight and a source belonging to it. It is then finitedimensional and irreducible, Proof The weights as points of a lattice can nowhere accumulate. By the

inequality 42.4 they are bounded. Thus their number is finite. By 42.2.2 their multiplicities are finite. By 42.2.1, R is spanned by weight vectors. Thus dim R < co. Let S be sub R and invariant, S # (0).One still has to show that S = R. Since dim S < m, there is a y E S belonging to a weight p, which is supposed to be maximal with respect to this property. Since e, S c S, e,y=O

for all a E W+.

The given representation of G extends to the associative envelope, which is mapped into End R.The Casimir tool (40.3) is mapped into

Now

zy =

C

acW+

(e,

e-,

+ e-,

e,) y

+ CJ hJh, y

43.

FINITE-DIMENSIONAL IRREDUCIBLE REPRESENTATIONS

22 1

Let x be a source belonging to the top weight A. Then there is a u in the envelope of G such that y

= ux,

and by the same argument as above zx = ((A

+ 6, A + 6) - (6,s)) x.

Because of zu = uz (see 40.2), zy = zux

(A

= uzx, it

follows

+ 6, A + 6) = ( p + 6, p + S),

thus by 42.4 p

=

A.

Because of Proposition 42.2.2(1)y is a scalar multiple of x; thus S = R, which proves the assertion.

43. FINITE-DIMENSIONAL IRREDUCIBLE REPRESENTATIONS

The notations are as before. Results are summarized and a few related ones are added. 43.1. Theorem Let the linear representation in R E Spa Lin Com (dim R < a)of G E Alg Lie Com S S be irreducible with the top weight 1in ordered second dressing. Then 43.1.1 R is spanned by weight vectors. 43.1.2 The weights belong to HLp. 43.1.3 The length of an u-ladder of weights starting at A is -2

(A2) * (a,4

43.1.4 S, turns the a-ladders upside down. 43.1.5 Int W* maps weights into weights. 43.1.6 The top weight is dominant and has multiplicity 1. 43.1.7 Int W*-equivalent weights have the same multiplicity. 43.1.8 The maximum of (A weight 1only.

+ 6, A + 6) for weights A is attained by the top

222

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

43.1.9 The image of the Casimir tool under the representation is the scalar multiplier (2 6, 2 6) - (6,6), where 2 is the top weight.

+

+

43.1.10 The set of weights is characterized by the following statement: it contains the top weight and for any p E W + + if , it contains p, also contains all mod p congruent elements of H: between p and Spp. 43.1.11 Within the closure of the dominant chamber the (dominant) weights p are characterized by the following condition: 2 - p is a sum of members of W ; otherwise stated : p G 2 in the minimal partial order. ++

+

43.1.12 e,e-,x = $ ( j l)(p -j)(a,a)x if x belongs to an a-ladder, in which it occupies the ( j + 1)th place and is followed up by p - j elements. Proof (1) See 42.2.1; (2) see 41.4; (3) see 41.3.3; (4) see 41.3.5; (5) con-

sequence of43.1.4; (6) see 42.2.2 and note that by 43.1.5 the Int W *-equivalents of $ are weights, thus lower than or equal to 2. (8) See 42.4. (9) By 40.2, the image of the Casimir tool commutes with all G ;by 36.3, applied to Lie algebras, it is a scalar multiplier; its actual value was computed in 42.5, when it was applied to the top weight vector. (10) Consequence of 41.3.5; (12) see 41.3.4; (7) and (1 I ) are left to be proved. As to (7), S E Int W* is induced by an inner automorphism of G (see 33. I), say c?, leaving H invariant. According to 41.1.2, c maps the weight space of X into that of Y h X(C"-'h) = SX. By reversing this argument one concludes that this mapping is onto. Therefore both eigenspaces have the same dimension. Remark Any u

E Aut

R that normalizes G and H induces Yxecuxu-'

E Aut

G

and

Aut w*. It is still true that this kind of automorphism preserves the multiplicity of weights. As to (1 l), according to the proposition of 42.1, every weight fulfills the condition. Suppose that ( is dominant and that [ + C p , p j ( p , E Wtt,pj integer 2 0) is a weight. One must ascertain that 5 is also a weight. It may be supposed that C p i p i # 0. Since YfEH.,.

YhEHS(UhU-') E

(C pipi, C PIPI) '09 there is a j such that

PJ>O,

( C ~ i ~ i , p j ) > O ,

43. FINITE-DIMENSIONAL IRREDUCIBLE REPRESENTATIONS

223

Hence

(5 + C P i P i , P j ) = ( 6 , p j ) + (CPiPi,pj) > 0, since 6 is dominant. Now by 43.1 .lo,

t+c

PtPl

-PI

is a weight. Repeating the argument one can descend to 6, which turns out to be a weight. This completes the proof of 43.1. 43.2. An Alternative Approach Theorem Under the conditions of Theorem 43.1the weight system does not depend essentially on the choice of the trunk. An automorphism T of G which leaves the trunk invariant turns the top weight 3 off into the top weight of Y,,f(T-' a) up to Int W*-equivalence. and 33.5. If the appeal to the group This follows from 41.1.4,42.2.3,33.4.3, G ISto be avoided (see41.1.3),one must rely on the weak structure of the weight system which has to include information on multiplicities and tell whether the difference between two weights is a rootform (see 41.2). These data determine the ladders and their lengths which, thanks to the ladder-length formula, locate the weights with respect to the rootforms and consequently fix the weight system within a linear space. In the proof of 43.1.7another appeal was made to G. This proof can also be modified to rest on the more elementary fact that S extends to an automorphism T of G (rather than to an inner automorphism). It is still true that the A-weight space of the representation Y,,f(T-'1;>equals the SA-weight space of the originalf. In particular, by 43.1.5both have the same set of weights and consequently the same top weight. In Section 44 finite-dimensional irreducible representations with the same top weight turn out to be equivalent by means of an equivalence that maps weight spaces belonging to the sa,me weight onto each other. This being taken for granted, it turns out that the A-weight space of the original representation is mapped onto that of the representation Y,,f(T-' a), which in turn coincides with the SA-weightspace of the original one. So the A-weight space and the SA-weight space of the original representation have the same dimension, which proves the assertion..

43.3-6. The Value of the Casimir Tool 43.3 If G E Alg Lie Com SSS, then the adjoint representation ad is irreducible. The weights are the rootforms; the top weight is the top rootform b.

224

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

The image ad z of the Casimir tool under the adjoint mapping ad, according to the notation of 40.1, is ad z = it&. 1

Because of the definition of the correlate bases, one gets

tr ad z = dim G, and because ad z is a scalar multiplier in G adz=l. With 43.1.9, this leads to the following. Proposition The adjoint image of the Casimir tool is the identity. = 1 for the top rootform d if G is simple. (If C is not simple, the proposition still holds for every simple component of G.)

(a + 6, 8. + 8) - (6,6)

43.4 G E Alg Lie Lin Com SS, G c End R, R E Spa Lin Com, dim R < a. tr(ab) as a (symmetric) function of a,b E G is an infinitesimal invariant of the adjoint Lie algebra since

e,

tr ((?a)b + a(Eb))= tr (cab - acb

+ acb - abc) = 0.

If, moreover, G is simple, thus d irreducible, it has to be a constant multiple of the Killing form: tr (ab) = K tr (ah).

43.4.1

For later use it is important to know: 43.4.2. Proposition If G is simple, K

=- 0. (Proof in 43.6.)

43.5 Let linear simple semisimple G be acting in R and let zo and z be its Casimir tool considered as acting on R and G, respectively, in agreement with the extension of C to its associative envelope &(G).It then follows from 43.4.1 that 43.5.1

trzo=KtrzN.

Let 3 be the top weight of irreducible simple semisimple C, and d the top weight of the adjoint representation, that is, the top rootform. Then by 43.1.9 43.5.2

tr zo = dim R

- (0+ 6, 2 + 6) - (6, a)),

44. 43.5.3

225

CONSTRUCTION OF REPRESENTATIONS

+

tr z N =dim G . ( ( 8 6, B + 6) -(&a))

= dim

G

because of 43.3. From 43.5.2-3 and 43.5.1 the following obtains.

+

43.6. Proposition K = ((dim R)/(dim G)) ((A + 6,A 6) - (a,&)) if simple semisimple G is irreducibly represented in R (dim R > 1) with top weight 2. Since (3,6) > 0 and (A,$ > 0, (1+ 6,A + 6) - (6,s) > 0. If this is applied to the irreducible components of the identical representation, one gets Proposition 43.4.2.

44. THE CONSTRUCTION OF ALL FINITE-DIMENSIONAL REPRESENTATIONS

The notations are as before. 44.1 Thanks to the conducibility theorem, to know all representations of G E Alg Lie Lin Corn SS in R E Spa Lin Corn (dim R < co), one can restrict oneself to the irreducible ones. These are exhausted by the following theorem. Theorem Up to equivalence a finite-dimensional irreducible linear representation of semisimpleGin ordered second dressing is characterized by its top weight. Possible top weights are just the dominant integral elements of H z . Remark This stresses again that different integral elements of the dominant chamber cannot be Int W*-equivalent. The same was shown, however, in 33.8.1 without assuming iutegrality. The first part of the theorem is proved in 44.6, the second in 44.2-5.

44.2 The second part is proved by an explicit construction of a finite-

dimensional irreducible representation from a dominant X play the role of top weight. The vector x belonging to X must be annihilated by all em(u E W + ) ,all b - X(b) (b E H), and all e!f'

E H&

which has to

t;3

( p E W + + p, - p = 2-

This suggests a construction of R as a linear image of the associative envelope €(G)of G ;the role of the 0 in R then has to be played by the left ideal M of &(G)generated by the elements mentioned above. The construction proceeds in two steps. First one uses the left ideal M' of b(G) generated by eOr (a E W + ) , b-X(b) (b EH),

226

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

and defines

f’ = Y a€C Y

U+M’,UEQ(G)

(au

+ MI)*

Then for every u E G,f’(u) is a linear mapping of R’ = 8(G)mod M ’ into itself andf’ is a linear representation of G in R’. The following conclusions seem to be obvious: 44.2.1 x’ = 1

+ M ’is a source off’,

44.2.2 x’ belongs to the weight h since all h - h(h) E M ’ , 44.2.3 A is the top weight because x‘ is annihilated by all e, (a E W+).

The conclusion that h is a weight, however, may be drawn only if x’ # 0. This rests on proving that -1 4 M’, which is done in 44.3. 44.3 Assume 44.3.1

where the h, are running through a basis of H and the u,,v, are taken from

8(G). This assumption has to be refuted. The actual construction of b(G) in Section 39 has to be taken under consideration. One may assume that the ordered basis X of G consisted of all e-, (a E W+),all e, (a E W+),and all h,

in this arrangement.All terms in 44.3.1 are now assumed as belonging to d ( X ) , and 44.3.1 itself is considered as a congruence mod / ( C ) . Furthermore, u,, Y, are assumed to be orderly. Thanks to Theorem 39.3, the decision whether 44.3.1 is possible is easy when the right-hand member of 44.3.1 is orderly. In the present form, however it is not. It has still to be refashioned mod /(C).This can be done as follows: u, is a sum of orderly monomials u’. In every u‘e, the e, is moved mod $(C) stepby step to the left, first through a few hi,then through a few eS (/3 E W + ) .The by-product of any step is a shorter monomial with at least one branch belonging to a positive rootform as a factor. Fortunately, e, need not change only slight rearrangements are required. places with any e-8. In v,(h, - A@,) Finally, the right-hand member of 44.3.1 will have been made orderly. With the original right-hand member it still shares the property that it will become 0 on substitution of 0 for all e, (a E W’) and A@,) for all h,. By the

44. CONSTRUCTION OF REPRESENTATIONS

227

uniqueness of the orderly element in each left coset mod f ( C ) the new righthand member has to be -1. This is a contradiction. 44.4 Thoughf' fulfills 44.2.1-3, it need not be irreducible; R' has still to be reduced mod an &(G)-invariant subspace T. Abbreviating the notation by writing u' instead off'(a) for any a E &(G),

one defines T as the smallest &(G)-invariant subspace of R' containing all 44.4.1

The representationfinduced in R = R' mod Tis limited. Thanks to 44.2.1-3, it fulfillsthe supposition of Theorem 42.5. Therefore it is finite-dimensionaland irreducible. Again, to be sure that it stillhas a vector of weight A, one must show that x' # T. T is spanned by the weight vectors 44.4.2

u'

eL$-P+I

where u ' = e L p I . . .el_pa e'm i

XI,

.. .e'

Ub

h ' k l . . .h;kI 1

may be assumed such that p I , .. ., p a , u,, . .., a, stem from W++in an arbitrary succession with possible repetitions. Here the h' can be dismissed because they are acting like scalars. x' can belong to T only if there are vectors 44.4.2 of the same weight as x ' . This can happen only if b 2 1. It will be shown, however, that 44.4.3

e' e'p-p+l x' = 0, a -P

which then makes x' E Timpossible. To ascertain 44.4.3, e: is moved to the right. If o # p, then o - p # W, [e&eLp] = 0; therefore e: passes readily through the eLp and finally annihilates x' which is a top weight vector. If u = p, then e'P-p+'X' =0 P -P

because of 41.3.1 and the choice ofp-,. (Note that this formula must be applied with p-p instead of i, and p instead of a,and that it does not depend on an assumption of finiteness of ladders.) This completes the proof of the second part of Theorem 44.1. 44.5 Reconsidering the construction of the representation, one may state the following.

228

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

Proposition G is irreducibly represented by Y a s c YU+M,UEB(G)

(au

+ MI,

with the top weight A in the linear space &(G) mod M , where M is the left ideal generated by the e,

h - A@)

(m E W + ) ,

(h E H ) ,

e!;P+'

(p

E

w++,p-p= 2"').

(P, PI

44.6 To prove the first part of Theorem 44.1, one has to consider another irreducible representation f * of G in a space R* (dim R* < w), with the same top weight A and the A-weight vector x*.f * extends to a homomorphism of b(G) into End R*. It must be brought into equivalence withfrepresented in R with the A-weight vector x.

f*(e,) x* = f * ( h - A(h))x*

for u E W+, h E H

=0

because x* belongs to the top weight A. Thusf *(M')x* = {0}, which implies that f*(M)x* is spanned by

f *(up)f *(e!;p+')x*

(p

E

W++),

which all vanish because f * is finite-dimensional and the length of the (-p)ladder starting at A is just p-p. Therefore,

f*(M)x* = (0). Now rp(u

+ M)=f*(u) x*

defines a linear mapping rp of R into R*.

+M)

rpf ( a ) ( u

= rp(au

+ M ) = (f*(au))x* =f *(a) f *(u)x* =f*(a)rp(u + M ) ;

hence p?f(a)=f*(a)rp, which showsf and f * to be enchained by means of rp and therefore equivalent according to 36.2. Finally, it is evident that equivalent representations actually have the same top weight on any ordered trunk. This completes the proof of Theorem 44.1. 44.7. Proposition The set of all weights of all finite-dimensional linear representations of G is H&. Indeed, these weights belong to H E g ;on the other hand, every element of HEo is Int W*-equivalent to a dominant one, which by 44.1 is a weight of a

45.

229

THE FUNDAMENTAL WEIGHTS

finite-dimensional linear representation. By 43.1.5 the same is true of the given element of H&.

45. THE FUNDAMENTAL WEIGHTS

The notations are as before. 45.1. Definition The T, E H:, defined by 2-=( (Ti,p,)

1

for i = j , for i Z j ,

0 where p,, . .., pl form the natural basis W + +of If:, are called thefundamental weights. The irreducible representations with these as top weights are also called fundamental. They are denoted correspondingly byfJ or simply by T,. The T, are dominant and integral and therefore really are possible top weights. Clearly, every possible top weight X is in a unique way a sum of fundamental weights, namely, (p,,p,)

45.1.1

Note that C r, = 6. 45.2 Linear representations f , g of a group G in R , S E Spa Lin produce a new one, f Q g in R 6 S, by Kronecker (or tensor) multiplication. On bases x,,xz, ... of R,y,,y,, . .. of 3, xi 6y,, i, j = 1,2, . . . of R 6 S, one defines

€3 Y,) = ( f ( 4 X i ) €3 (g@)Y,) For Lie algebras G this implies the following

((f6g)(a))( X i

for a E G .

((f Q g)(a))(xi 6 Y,) = (f(a)xi) 6 Y J + xi 6 (g(a)Y,). Kronecker or tensor products with more factors are analogously defined. f 6 g is readily seen to be a linear representation of G in R €3 S iff, g are so in R , S . Definition

45.3 For finite-dimensionalfand g, if xI,y, belong to the weights hi,p,, then x i 0y , belongs to A, p,, which consequently is a weight off Q g. Note, however, that the irreducibility offand g does not imply that off 6 g. Because of the conducibility, an element belonging to a top weight off €3 g is still the source of an irreducible linear representation with this top weight. Therefore,

+

39-50.

230

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

x

if A is a possible top weight, its presentation as a sum p i n , of fundamental weights shows that an irreducible representation with top weight A can be recovered in a Kronecker product ofp, factors n, (i = 1, ..., I ) . Theorem The finite-dimensional irreducible linear representations of C E Alg Lie Com SS can be obtained from the fundamental ones by Kronecker multiplication (45.2) and isolating the linear subspace a source of which is a

vector belonging to the top weight. 45.4 The weights, top weights, and fundamental weights of a direct sum G = GI G2 split according to the splitting of the trunk H = HI + H2into components that are weights, top weights, and fundamental weights of the components. Therefore their knowledge is reduced to that for simple semisimple Lie algebras.

+

45.5 A linear representation of a group Gin a finite-dimensionallinear space R admits a dual one in the dual space R* of R :

f * = Y a e G Yu€R* Yx€Ruf(u-')x. This leads to the following definition. Definition The dual of a linear representationf of G E Alg Lie in R is

f *=

YOCG

Y u € R * Y x € R (-'f(a)x)*

Proposition Dual representationsof semisimpleLie algebrashave opposite weight systems. This is easily shown with the use of a basis of weight vectors. In the particular case G E A,, up to equivalence effectuated by a basis choice,

f * =f% where 7 is the outer automorphism of 24.6,q = Ya(-a'). 45.6. Integral Forms, Dominant Elements, and Fundamental Weights and Representations for G E A,, B,, C,, D,,in the terminology and with the natural bases found in Section 16 (see also 25.7 and Table C). A, : the w, are linear functions on H subject to For 5 E H* the coefficientsp , in

x w,(h)

= 0.

45.

THE FUNDAMENTAL WEIGHTS

23 1

are determined up to a common summand. It is more convenient to suppose that they are “reduced,” that is, such that

CP,=O. The inner product of

5 = c P,W,Y

rl=

c q,w,

is defined by

(5, 7) = X P j q j as soon as one of the factors is presented in reduced form. The rootform denotations w i - w, are that kind. Hence for i = j -1 for li-jl = 1 (0, - ~ t + l w r j - w j + l )= for li-jl > 1.

1:

Up to a common factor (indicated in Table C) this is just what is required in second dressing. Integral respectively dominant elements 5 = C p , w j E Hz have to make

(Z Pi wi,w, - w,+J (w, - W,+IY w j - w,+J

integral respectively

2 0 which means that

p , = p , mod 1,

respectively p , > p , + ]

( j = 1, ..., I).

The fundamental weights (now in an unreduced presentation) are TI=

w1

7r2 = 0 1

7r,=

w1

+

w2

+ w* + *..+ w , .

In its usual representation (see Section 16) by the special linear group of

( I + 1)-space the ith basis vector belongs to w i . The top weight is just w 1 = 7 r l .

To construct the next fundamental representation, one takes two copies of (I+ 1)-space with bases xl, ..., xi+, and y , , ...,yl+,, and acted on by G E A,. Their Kronecker product splits as a direct sum of two invariant subspaces, the symmetric part with basis elements x , @ y , + x, @ y , (i ~ jand) the skew part with x i @ y , - x, @ y , ( i cj).For the symmetric part the weights are the w , + w , (i ~ jwith ) 2 w , as the top weight; for the skew part they are w , + w, ( i <j ) with w 1 + w2 as the top weight. Now it easily follows from 42.2.5 that both are irreducible. Thus 7r2 is the representation induced by 7r,(C)in the linear space R(2)of

232

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

skew 2-tensors of ( I + 1)-space R. It is closely connected with the manifold of straight lines in projective I-space (Plucker coordinates; see 25.8.4’). Similarly,rkcan be defined in the space R(k)of skew k-tensors of (I + 1)-space (connected with the (k - 1)-planes in projective Z-space). RCk)and R(l+l-k)are in a natural way duals of each other. ?rk is easily seen to be equivalent to - T I + I - k . Therefore by 45.5, up to equivalence, the representations T k and T,+l-k are duals of each other. In other words, the automorphism )‘,(-a’) interchanges T k and Tl+I-k* B,: The I functions w, on H a r e given the inner product (w1,

=

(01

for i=j, for i # j .

The elements of W + +in the enumeration of 25.7 are w l , w I- w 2 , ..., - w,, which again verify the conditions of second dressing up to a common factor (to be found in Table C). wl-l

Integral 5 = p,w, has to fulfill: 2p,,pJ -P,+~integral. Dominance requires : p , > p,+, > 0. Fundamental weights :

+ - - + w1)

=

*

T2 = W I

T J= w1+ w2

TI = w1

+ w2 + ’ ’ + w1-1. *

to the usual presentation of G (see Section 16) by means of the special orthogonal group of (21 + 1)-space. The various T k (k > 2) are constructed, as in the case of A,, with an analogous geometric interpretation This point will be elaborated in 70.1-2. The meaning of v I is given in Section 49. Note that here the symmetric part of the tensor product of r2with itself is not irreducible. The invariant quadratic form produces a one-dimensional invariant linear subspace.

n2 belongs

C, : Inner product as for B,.

w++:

w1

- w2, . . ., w1-1

Integral 5 = 2 p,w,: p j integral. Dominance: p j > p i + l > 0.

- w,,2w,.

45.

233

THE FUNDAMENTAL WEIGHTS

Fundamental weights : TI = w1

T,=w] +w, T,= W I

+ w2 +

* * *

+ w,.

Interpretation as for A,. Note, however, that here the skew part of the tensor product of n1with itself is not irreducible. The invariant skew form produces a one-dimensional invariant linear subspace. In general, the linear space of skew k-tensors need not be irreducible.

D,: Inner product as for B,.

w++:w,-1 - a,, w1-1 + w,+1, Integral

w1 - w,,

w2

- w3,

. a

., w,-2

- W,-l.

e= x p j w j : p j -p,+I,pl-l + p l integral, that is, 2p, = p i - p ,

Dominance: pI > p z > Fundamental weights :

-

* -

TI = +(w1

mod 1.

=0

>pI-l > lpll.

+ + w2

* * *

+ w,-1 - w , )

"'+d,-,+w,)

q=+(w1+w2+ T3 = w I ~ 4 = ~ 1 + 0 2

T,= w ]

The interpretation of T Section 49.

+ w2 +

. ..,

* *

+

~ , T,causes no

01-2.

trouble;

will be explained in

7rl, T,

45.7 The fundamental weights on a natural basis for all G E Alg Lie Com SSS are to be found in Table F. 45.8. Historical Note t o Sections 41-45 The theory of irreducible linear representations of complex semisimple Lie algebras with notions like top weight goes back to E. Cartan, Bull. SOC.Math. France 41,53-96 (1913) = (Euures I , 1, 355-398, though his proofs for general theorems like those corresponding to 44.1 are not convincing. The present method was derived from Harish-Chandra's Trans. Amer. Math. SOC.70, 28-96 (1951). His approach

234

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

has been simplified mainly by the use of the inequality 42.4 which has proved to be of primordial importance (H. Freudenthal, Proc. Kon. Akad. Wet. Amsterdam ASI, 369-376 (1954) = Indagationes Math. 16). It should be noticed however, that Harish-Chandra’s method can also be used for an a priori construction of the complex semisimple Lie algebras themselves.

46. THE FUNDAMENTAL GROUP OF UNITARILY RESTRICTED SEMISIMPLE LIE GROUPS

G E Gru Lie Lin Coin S S , in ordered second dressing; LIPto 46.4 centerfree. The problem dealt with in Section 32 is 46.1-4. Reenvisaged 46.1 In 32.2 the fundamental group the set Z c b c Hun:

Qj

of Gun(and of C ) was described by

h E Z t , h E DA a(h) = 0 or 2ri for every a

E

W+.

The straight path from 0 to h E Z is mapped by exp into a closed path on C, a representative of a member of 0.The fundamental group operation could be performed as an addition on the elements of 2; to stay in Z or rather to bring the sum back into Z one had to use the operations of the kaleidoscope group Int W* which do not change the homotopy class of a path. This is an unsatisfactory description, which will now be reshaped. First, the set Z i s extended to the corner lattice (33.14.3):

h E 2- a(h) = 0 mod 27ri

(a

E

W*).

Again exp maps the straight paths (cf. 31.4 for terminology) from 0 to h E Z into closed paths on G . How to find the homotopy equivalences among them? Some of them are known by the fact that inner automorphisms do not change the homotopy class. This information will be sufficient. Denote by 8 the canonical homomorphism YhEZ[Yt.OstsL

exp fh1

of the additive group Zonto 0, and by Y its kernel. Now, as mentioned above, Int W* exerts no influence on @, that is,

SSah=9h

for h E Z ,

EE

W*.

Thus 2ri * 2a(h)ha/(a,a) E Y, which for a E W++and suitable h E Z implies 2ri*2h,/(a,a) E Y for a E W++,hence by Proposition 41.5 for all E W * . Therefore, if Sa,,,is the reflection in the kernel of a - m 27ri for a E W * and m an integer

46.

FUNDAMENTAL GROUP OF UNITARILY RESTRICTED GROUPS

235

is in Y because it is an integral multiple of 2h,/(a,a) .2ni, since a(h) = 0 mod 27ri. By 33.14.6, any h E Z can be mapped by a composition of these reflections into a corner of D. On the set of corners of D 9 is one-to-one. So, if h E Y, it follows that h is an integral combination of the 2ha/(a,a)*277iwith CL E W*. The result is the following. Theorem For centerfree Gunlet Z b e the additive subgroup ofHunconsisting of the h with a(h) = 0 mod 277i for all a E W*, and let Y be its subgroup generated by the 27ri(2ha/(a,a)). Then 2 mod Y is isomorphic to the fundamental group @ of Gun by the following mapping: let w be a path from 0 to h E Z ; then exp w represents the element of @ that corresponds to h + Y. 46.2 The result can be reformulated by employing duality for abelian groups. Definition A cliaracter mod 27ri o f a (discrete) additive abelian group A is a homomorphism into the addition group of imaginary numbers mod 277i. The characters of A again form an abelian addition group denoted by A*, the dual of A , which, if topologized by pointwise convergence, is compact. If A is finite, A and A* are isomorphic. In the present case the dual of 2 mod Y is made up in a natural way by the characters mod 2ni of Z that vanish on Y . Let fz be a character mod 27ri of 2.Let 2, be an integral basis of Z. tz(Z0) is a set of imaginary numbers mod 277i. There is a 5 E H: such that ((z) = &(z) mod 277i in every point of 2,. Then f ( z ) = &(s) mod 27ri for any z E Z . Such a 5 is said to extend tz to Hun.This extension is not unique. Suppose now that tz vanishes on Y . For 5 this means

for all a E W*. = 0 mod 27ri (a, a) Since [(h,) = ([,a),this is equivalent to 5 E H&. To relate this to the dual of Z m o d Y one has to identify the elements 6 of If,*,,that behave the same way mod 2ni on 2. 5 behaving as 0 on Z means that ((z) = 0 mod 27ri whenever a(z) = 0 mod 277i for all a E W*. In other words, 5 is an integral linear combination from W*. This suggests the following. 2ni * 2

~

46.3. Definition Put A = HI",,= the addition group of the set of all possible weights of G and, iffis a finite-dimensional linear representation of G,

call Af the (discrete abelian) subgroup of A generated by the weights off. In particular, A , is the group generated by W * . The result proved now reads as follows.

236

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

Theorem The fundamental group @ of centerfree Gun is the dual of A mod A , in the way that for any exp w E 'p E @, where w is a path on Hun

starting at 0, the character mod 2ni Y,(E(w(l>) mod 2 4

of@hasto beidentified with the element f

+A ,

ofA mod A , .

Remark Note that the identical action of Int W* on @ is reflected in the identical action of Int W" on A mod A , ; the latter property is a direct consequence of the definitions of A , A % ,and Int W*.

46.4 In computations of fundamental groups (or rather their duals) a convenient basis of A consists of the fundamental weights (see Table F). Notation Z,(n) means thep-cyclic group with the generator n

+ A %.

An easy computation leads to the following list of homotopic relations (indicated by -) and descriptions of the

-- - -- +- -- -

Duals of the Fundamental Groups of Centerfree Gun of G E

A,:

r k

knl,(I+ 1)nl -0:

Zi+I(rl).

-

r l4 0 : Z2(n1).

8 , : 2nl

n2

c,:

0, ?rk m knl for k < 1, r 1

2nl

D,: n k - 0

2r3 277, 2nl E6:

7T2

z2(7r1).

7,:

foreven k > 2 , T k mh3 forodd k > 1, 0, 2 r 1 2n2.Furthermore, n3,r 1 7r2- 0 for odd I: Z4(7rl), 2n2 0, r3 n 1+ 7r2 for even I : Z 2 ( n l + ) Z2(7r2).

71@ T 5 , T 3

*r E, : trivial. E7

* * *

3

7 4

C/r

----

2T1, 37Tl

T6,27, mT I

@

T4

0, T 2 rs

TG

"7

0 : Z,(n,).

0:

z2(~2).

F, : trivial. G, : trivial.

46.5-6. Fundamental Group of a Representation 46.5 The assumption that G is centerfree is now dropped. The infinitesimal algebra G of G can be considered as the image of under a linear representation$ Then G is the linear Lie group infinitesimally generated by the linear

46.

FUNDAMENTAL GROUP OF UNITARILYRESTRICTEDGROUPS

237

Lie algebra ,fG. If Gi (i = 1, 2) are isomorphic, and their Giare identified according to this isomorphism (= C), then, in general, different fi may be needed to generate the Gi from the linear Lie algebrasfi G. However, under these conditions: 46.5.1. Proposition The set A, is the same for isomorphic Gun. Indeed, it is characterized by for all B with expf(k) = I , X E A, tt X(B) = 0 mod 27ri since the weights offgenerating A, are just the eigenvalues on H .

Gun wraps Gun by means of the inverse of f. Those closed paths exp tk with h E Z that are wrapped by closed paths are characterized by expf(k) = 1, hence by X(B) = 0 mod 27ri for all X E A,. This leads to an extension of Theorem 46.1 :

yz,,

46.5.2. Theorem For Guninfinitesimally generated byf(&,) let Z, be the additive subgroup of Hun consisting of the B with X(h) = 0 mod 27ri for all A E Af.Then Z, mod Y is isomorphic with the fundamental group of Gunas in Theorem 46.1.

Translated by duality this becomes the following. 46.5.3. Theorem The fundamental group @ of a unitarily restricted semisimple linear Lie group Gunis the dual of A mod A, if G is produced by The duality has essentially the same meaning as in Theorem 46.3; that is, with the provision that exp fw replaces exp w and that A, replaces A .The center of Gunis isomorphic to the dual of A, mod (1-. The last statement is an immediate consequence of the foregoing.

fe.

-

46.6. Proposition If in the foregoingfis irreducible with the top weight 1, then the center of the group infinitesimally generated byfcun is p-cyclic, where p is the least positive integer such that p 2 E A N . Indeed, then all weights offare contained in 1 A (see 42.1).

+

46.7-9. Wrappings Realized 46.7 The universal wrapping of Gunhas a center that is isomorphic with the fundamental group of Gun.Therefore it cannot be realized by an irreducible unitarily restricted semisimple Lie group unless the fundamental group of is cyclic.

e

238

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

In the case D, (I even) it can be done by combining representation.

into a reducible

rl,7r2

46.8. Theorem Any wrapping of the centerfree unitarily restricted semisimple Guncan be realized by a linear Lie group. Proof The wrapping is characterized by its fundamental group as a subgroup of the fundamental group of Gun;its dual is isomorphic to A’ mod A N , where A’isa subgroup of A containing A . There is a (finite) set A” of representatives of A’ mod A -, all of which are dominant (cf. Remark 46.3). A’ = A ” + A -. The representations with top weights in A”can becbmbined to give a representationf, for which A, = A’.

46.9 According to 38.5 the fundamental group of complex semisimple G is known by that of its unitary restriction. 46.10 Another Approach If rather than the fundamental group of Gun (generated by its center were the aim, one could proceed in a simpler way than that followed in the present exposition. This way can be sketched in a few words.

f(eun)),

(1) The center of Gunis contained in the trunk Hun. (2) Its elements exp h are characterized by exp h = 1 ; in other words, by a(h) = 0 mod 27ri for all a E W *;that is by X(h) = 0 mod 237i for all A E A H , (3) exp h = 1 if and only if A@) = 0 mod 27ri for all X E A,. (4) From this it follows that the center is isomorphic to the dual of A, m o d A N .

This method, however, sets no upper bound, as it were, for the fundamental group. To know that the fundamental group of Gunis finite, that the universal wrapping of Guncan be realized by a linear and even by a unitary Lie group, and so on, one must, to the present knowledge, apply more profound methods that is, homotopy theory, as has been done in this exposition, or homology theory.

47. WEYL‘S CHARACTER AND DIMENSION FORMULA 47.1-6. Skew Functions and Characters 47.1 The double characterization up to equivalence of irreducible representations of G and Gunby their top weights and characters suggests the possibility of expressing the one by the other.

47.

WEYL'S CHARACTER AND DIMENSION FORMULA

239

Until further notice Gun is supposed simply connected. Without this supposition one would not be sure whether a given linear representation f of G extends to G. By 46.8, this supposition is allowed. In any case G is taken in ordered second dressing. 47.2 A character of Gunis known by its behavior on Hun.One can work more comfortably in H:n, defining by

x,

47.2.1

= tr

X,(Ch)

expfh

(for the mapping 5 of H onto H* cf. 2 1 . 1 3 . For 7 = ( h the eigenvalues offh are the weight values A(h) = (A, T ) , each of which has a multiplicity

mA= dim RA,

47.2.2

where RA is the A-weight space. The eigenvalues of expfh are the exp(A,~); thus,

For S

E Int

W*

~ , ( S T )= C mh exp(A, ST)= h

A

mA exp(S-' A, 7 ) = C msAexp(A, 7 ) ;

thus by 43.1.7,

47.3 In addition to these functions invariant under Int W * , it is convenient defined by to consider skew functions 0,

@(ST)= det S O(7)

for all S E Int W * ;

in particular, elementary skew functions O,, defined for p

C det S exp(Sp, 7 ) .

0,(7) =

47.3.1

E H::

*

s6lntV

Note that for S

E

Int W* Osr(7)= det S * 0,(7).

Let p be integral and dominant. If (p,p) = 0 for some p E W++,then OJ7)

=

Os,, ( 7 ) = -0,(7).

Thus 0,(7) = O

for all

7

E H:,,

240

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

unless

But then, using the Weyl tool S, 2'cQ)

2

o

for all p

(P,P)

E

w++

and p - S is still dominant. Thus: Proposition The only nonvanishing elementary skew functions 0, with integral dominant p are the Or+* with integral dominant v. 47.4 Definition

This definition differs slightly though inessentially from that of 37.3. According to 41.6, S ( E W++)permutes the elements of W + except p, P p which is interchanged with -p. Thus, Q ( ~ , T )= - Q ( T ) ,

which means by Proposition 33.2.2 that Q is skew. In working out the product one gets a sum of terms +exp(t, T ) . The highest 5 that occurs is $a = 6 (the Weyl tool); the others differ from S by an integral combination of positive rootforms. Q is a linear combination of elementary skew 0, with integral dominant p. The highest p that can occur is 6. Actually, it occurs with a unit coefficient. By 47.3 all 0, with a lower p vanish. Hence :

zaGw+

47.4.1. Proposition

Q(.)

=

WT).

47.5 Guided by 37.3, one is advised to replace x, by ,O 47.5.1

O,(T>

= x,(T)

Q(7).

The Of,as defined in 37.3, fulfill orthogonality relations in H . The new ones do not differ much, though there may be some doubt whether Q can still be transplanted to Hfor integration purposes; that is, whether Q( 2. The first member involves multiplicities of weights >A only and the coefficient c2 - cA in the second member is well known not to vanish (see 43.1.8). The formula is more practical than Weyl’s, which will be derived from it, or rather from 48.2. 48.4 The equation system 48.2 is now subjected to the Laplace transformation that consists in multiplication by exp(h,T) and summation over all

246

39-50. ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

X E A, or, rather, which amounts to the same, summation over all of HEp. This is how x(7) = mA

c

makes its appearance. If p is substituted for h +pa, the first summand contributes m

Now with dual,

5 as in 21.1.5 and canonically identifying H with its double linear exp(p, 7 ) p = grad, Y ,e x ~ ( p71, , exp(-(a' 5-l a = grad, y, log (l-exp(-(a, 1 - exp(-(., 7 ) ) exp(A, I Y T exp(h, s5-l

9('

T))),

L=

where the Laplace operator d = div grad is understood with respect to the inner product in H*. The result is 48.5

X

(grad Y,

acW*

1% (1 - ex~(-a,TI), grad x) + Ax = ch* x,

which can be simplified by the use of Q (see 47.4,47.4.1):

=

x det S exp(S8,

T),

S

Q(7I2 = f

Now 48.5 becomes '

n (1

asW*

- (ex~(-a,7 ) ) ) .

(2 grad log Q, grad x)

+ dx = c! x.

Because of grad log Q = (grad QYQ,

48.

ALGEBRAIC PROOF OF WEYL’S FORMULAS

this becomes 2(grad Q, grad

247

x) + QAx = c i Qx.

Putting Qx = 0, as in 37.3 and 47.5. I , one arrives at 48.6

(dO/Q- ( A QlQ)= c i a

The same presentation of Q (47.4.1) shows

d Q = (6,6) Q. Hence 48.7

de = (h + 6 , h + 6) 8.

0 is a linear combination of elementary skew O,,(T)= C det S * exp(Sp, T ) , S

with integral dominant p, all G h + 6. For any among them

( A 0,)(7) = C det S (Sp, Sp) exp(Sp, 7) S

= (p, p)

0,‘(7).

The linear independence of the nonzero 0, (a consequence of their orthogonality in H ) implies that (p,p) = 0 + 6 3 3 + 6) as soon as 0, occurs in the same expression for 0; dominant p G

h + 6 makes

p=h++* Since e x p ( h + 6 , ~ )occurs with a unit coefficient in both 0 and 0 ;

one gets

which proves Weyl’s formula. Note that the use of series and differentiations in this proof was merely formal. 48.8. Historical Note The contents of this section are based on H. Freudenthal, Proc. Kon. Akad. Wet. Amsterdam A51, 369-376 (1954) = hdagationes Math. 16. Examples of the practical use of formula 48.3 were given by H. Freudenthal, Proc. Kon. Akad. Wet. Amsterdam A51, 487-491 (1954); A59, 51 1-514 (1956).

248

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

49. CLIFFORD ALGEBRAS AND SPIN REPRESENTATIONS t

The fundamental representations of G E B,, D, with half-integral weights and n2)are still waiting for realizations. They are called spin representations. Since the usual presentation of G (n2,respectively r 3 ) has a center of order 2, the spin representations give I-Zimages of the special orthogonal groups. Clifford algebras are involved in their construction. (n,, respectively, nl

49.1-3. The Infinitesimal Approach 49.1 The special orthogonal group G of n-space is now taken with respect to

an invariant unit form in matrix presentation. Its infinitesimalelements are the skew matrices. For distinct U,b,d,,, means the n-n-matrix with 1 in the place ‘a,61, -1 in the place %,a1, and 0 elsewhere. The d,,,b(a c b) form a basis of G. Note that da,b = -db,,,. Ida,b, dc.dl =

if a # c, d ; b # c, d,

[da.b, db,c]= da,c

if a # b, b # C,

C#

a.

These relations determine the Lie algebra algebraically. Put 1 = [n/2];

a’ denotes the subscript a + 1 if 0 c a Q I and a - 1 if 1 < a G 21. The ordered trunk His the linear space spanned by da,+(0 < a G I ) as a basis, partially ordered according to the order of the subscripts a. On this basis the d,,,-coordinate of h is denoted by ~ , , ( h )The . branches and rootforms are da,b

da.b

- da*,b, fid,,,b,f id,,*,b belonging to + d,,,,be - id,,.b’ + id,,i,b belongingto

fi(Wo

+ Wb),

i(W, - mb),

and for odd n, in addition, belonging to +iua

da,nf

(a # b, a G 1, b G I ) ; the signs are understood to be correspondingly dependent. 49.2 The expression f,2

+

*

- + *

fn2

can be factorized into

(41P I + * * *

+ fnpn)’

by the introduction ofp,, fulfilling

t The contents of Section 49 are not used in the sequel.

49. CLIFFORD ALGEBRAS A N D SPIN REPRESENTATIONS

249

These pa are considered to generate an associative algebra with I-elemellti the Cliffordalgebra C,,, with the products 49.2.2

constituting a basis of C,,. The existence of C,,is proved by induction. Let C,, be an associative algebra with the basis 49.2.2 and subject to 49.2.1. Suppose dashing defined by Pa =

-Pa,

which generates an involutory automorphism of C,,. The ordered pairs of elements of C,, are appointed as elements of C,,,,. Addition and scalar multiplication are defined as usual in direct sums of linear spaces; moreover, rU],u2’

*

ru1, u21 = ru, u1

-i = r-

ru 1 9

+ uz 62, 242 dl +

241

uz1,

u1, - i 2 ’ .

2

The usual laws, especially that of associativity, are easily verified. One can also verify that dashing is again an involutory automorphism. The identification of u with ru,O1 embeds C,, in C,,,,. ru,,u21 * ‘0,ll = ruz, u, 1,

r 0 , i i . ru 1,

2 i = rU- 2 , a 1 1 ,

-

r 0 , 1 1 =-ro,l’.

By putting ‘0,l l = pn+, and doubling the old basis by right multiplication by pn+,the induction step is completed. 49.3 The second-degree elements of C,form a Lie algebra G‘if [u, u] = uu - uu

is accepted as a definition. The linear mapping K determined by 49.3.1

Kda.b = *pap,,

(0< a < b < n)

is easily seen to be an isomorphism of G onto G‘.The left multiplications 49.3.2

La = Y ” 4 U

(4 E G?

cause a linear representation

L =YasGLa of G’in C,,. So LK is a linear representation of G in C,,.

250 Forq

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

= Kdaoaf = +pop,,

one gets q2 =

-a;

hence,

Lg2u = Lq2u = d u .

All eigenvalues of L, are h+i. So possible weights of the representation LK with respect to the trunk H a r e +i(*w, f *

* *

h w,),

for which weight vectors, depending on the signs, are chosen: (D,) u ~ u $ * u*f ,*

49.3.3

with u: = p j

+ ip,,,

u7 = 1

(6,)

U : U ~ * - - UI' U 0 ,

+ ip,p,.,

uo = 1 + p n .

It is easily seen that left multiplication byp,(a = 1, . .., n) permutes them up to factors f l , 4. So they span a G'-invariant subspace R of C, of dimension 2', in which G' is represented according to r I for G E 6, and according to a (reducible) combination of r Iand r2for G E D,. It is easy to describe the action of the basic +L,,,, on R in a more explicit way by matrices on an eigenvector basis. This may be omitted. 49.4-6. The Group Approach 49.4 By the above device the spin representations of the Lie algebras G E B,, D, have been described. The result is not quite satisfactory. One would like to have the spin representations of the special orthogonal groups in an explicit form. To unify the exposition it is convenient to denote the universal wrapping of the special orthogonal group of n-space by G. This is a group locally isomorphic with a linear Lie group, as in 6.4. In an obvious way Gcan be understood as its infinitesimal algebra. Again,

I=[+n]

and

6=n-21=0,1.

G is represented in n-space by a 2-l-homomorphism 9 and in 2I-space by a l-l-homomorphism u. In the case 6=0, u is reducible and splits into r,and r 2 . To avoid half-integers the meaning of the w i is changed so that the nonvanishing rootforms are

f2wl f 2 w j f2wl f2w,

(i#j) for 6 =0, ( i # j ) , f2w, for 6 = 1.

The weights, all with multiplicity 1, are for

9: f2w1 for 6 = 0, supplemented by 0 for 6 = 1, u:

f w , f * . * f wp

If the eigenvalues of 8 ( x ) are e*2wr,supplemented by 1 for 6 = 1, then the

49.

CLIFFORD ALGEBRAS A N D SPIN REPRESENTATIONS

value of the character x of u for x

E

25 1

His the sum over all sign distributions:

defined up to a factor f l . The square of this character value is

where X ranges over all nonzero weights of 9, thus, det( 1 + 9(x)) = f det(1 + 9(x))

for 6 = 0, for 6 = 1.

=

Thus, first for x

E

H a n d then for general x,

49.4.1

x(x) = +(2-* det(1

+9 ( ~ ) ) ) ~ ' ~ ,

where the sign is to be settled by analytic continuation. 49.5 In the space of continuous functions on G a subspace K, spanned by the ) E G) or, equivalently, by the matrix coefficients of a, is functions Yx ~ ( x a (a singled out. Its dimension is (2')2 for 6 = 1 and 2(2f-')2 for 6 = 0 (see 49.3). Thus, dim K = 2"-'.

(An explicit basis of K is indicated in 49.6.) 49.6 As the next step the set of symbols 1, ..., n, or rather that of all its subsets with an even number of elements, is considered. This set of subsets is called N . It has 2"-' elements. For any v E N let e, be one of the two elements of G with the property

6(e,) is the diagonal matrix rAl, .. ., An1, with

A,= 1 for j $ v, X,=-1

for

~ E V .

(A fixed orthonormal basis was chosen in n-space.) If 9(x) = 8 ( y ) for x # y , it is convenient to write y = -x. Then u(-x) = -a(x) since the quotient of a(x) and a(-x) is of the form expo@) # 1 with h E H , 2w,(h) = 0 whence (*w l f f w,)(h)all equal ni mod 2ni. e,e,

=fe,

where p = ( p u v)\(p

n v).

252

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

By 49.4.1 x(eJ = 0 = k2'

if if

is nonvoid, v is void. v

The Yxx(xe,)are linearly independent because of X(e,ev) = 2'

for p = v for p f v.

E,

=O

(E,

= *I),

(See 49.4.1.) They form a basis of K.Hence for a E G, x(x4 =

z r u ( 4 x(xeu>,

YEN

where the coefficients r,,(a)can be found by putting x

r,(4 = 2-' E,

=

e,:

x(e, 4.

This leads to the formula x ( x 4 = 2-I

C

E,

uEN

x(ev4 x(xeu),

which remains valid if yxx(xa)is replaced by any linear combination

f(x) = Ca ma x ( x ~ ) of such functions; thus, f ( x ) = 2-'

c &uf(ev)x ( x e J

YEN

Tn particular, iffis any matrix coefficient of u, and even for u instead off,

~ (=42-' VCE N E, x(xe,) 49.6.1

where the square root involving the subscript v has to be interpreted as a continuous function on G which takes the value 2(*12)"E, at x = e,. 49.7. Reduction of Irrationalities Up to the precise knowledge of the

this is an explicit spin representation of the orthogonal group. However, it depends on 2"-' irrationalities while one square root is likely to do the job. This reduction is now performed. It is easily seen that for n-n-matrices a,b the relations

u(e,)

b = (1 - a)(l +a)-'

and

a = (1 - b)(l

+ b)-I

49.

CLIFFORD ALGEBRAS AND SPIN REPRESENTATIONS

253

imply each other (so long as they are meaningful) and that orthogonality of a implies skewness of 6, and conversely. A square root of thedeterminant of a skew 2k-matrix b is a rational function

the so-called pfaffian of 6, written pf(b), of the matrix coefficients /Ii,,

where sgn means the permutation sign. (The proof is omitted.) Suppose orthogonal a general enough to admit skew b = (1 - a)(l Then 49.7.1 det(1

+ a)-'.

+ a6(e,)) = det((1 + b) + (1 - b)6(eu))det(1 + b)-' det((1 + 6(e,)) + b(1 - 6(e,,)) det(1 + b)-'. =

To make the computation easier, one considers the case v = { 1, ..., m} with even m.Then 1 + 6(eu)has 0 on the main diagonal places 1,2, ..., m, and 2 on the other diagonal places, whereas in b(1 - 6(e,)) the first m columns of b are multiplied by 2 and the others are annihilated. Therefore the first factor in 49.7.1 is 2" times the determinant of the matrix b, formed by the first m rows and columns of b. 49.7.2

det(1 + a6(e,)) = 2" det b, det(1

+ b)-'.

To apply this to 49.6.1 one must substitute 6 ( x ) for a. Instead of b one writes ~ ( x )that ; is, 49.7.3

v(x) = (1 - 6(x))(1

+

S,(x))-l.

~ ( x is) skew. Its submatrix constituted by the rows and columns the numbers of which occur in v, is called q,(x). Then 49.6.1 becomes

where pf(7,) means 1 for void v. The doubtful signs are settled by a suitable choice of the e,. For i <j the curve Yrar on G is defined by a, = 1 and

7

7

1

254

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

where the cosines and sines are placed in the ith andjth rows and columns. Next there is defined e(t,,) = 4 ;

furthermore, if il < iz
, 0.

t The contents of Section 50 are not used in the sequel.

50.

251

CONDUCIBILITYTHEOREM, E. E. LEVI’S THEOREM

If tr zf

=

0,

thenfis a null representation. Proof On a suitable basis

with irreducible representations g , of G in R,. Thus, tr zf

=

C tr zgJ2 0,

according to 50.3.1. zgJis a scalar multiplier. If tr zf = 0, then tr zgJ= 0; hence zgJ= 0 for all j . By 50.3.1 every g, is an irreducible null representation; thus dim R, = 1. This makes f ( G ) , which is a homomorphic image of G, solvable. Consequently f(G>= (01. 50.3.3 Under the same supposition, with respect to the eigenvalues of z’, R is split into zf-invariant N , M sub R,

R=N+M, such that zf is nilpotent in N and nondegenerate in M. Proposition fleaves M and N invariant, and acts as a null representation in N. This follows from zff(a) =f(a)zf and from 50.3.2.

50.4 The Casimir tool is now generalized. Letf, g be linear mappings of semisimple G into End R, where R Com. Then, still,

E

Spa Lin

does not depend on the choice of the correlate bases, as is readily seen by rewriting the proof of 40.1, and again

39-50.

258

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

As soon as f becomes a representation, one can of course substitute [f(c),f(a,)] forf([c,a,]); likewise with g.

c

PJ = g(aJ)f(a,) = z g(a,)f(a’), J

since the bases may be interchanged. Thus by defining zcf*81 as

50.4.5

[f(c), zcf*pll= z [f(a’), [g(a,),g(c) -f(c)ll. I

50.5 To prove Theorem 50.1, consider a linear representationfof semisimple G in R with an invariant L sub R. There is an M sub R and a linear representation g of G in R such that

L + M = R, g(a) M c M , g(a) x =f(a) x

L nM for x

= {0},

E L,

(f(a) - g(a)) M c L.

(Iffis a matrix representation on a basis that extends a basis of L, thus

then g arises by “cleaningl’f, that is, omitting the contribution *.) The theorem actually states the equivalence offand g. It suffices to prove it under the assumptions fL and fM irreducible. The easy induction that leads from this case to the general one may be omitted. 50.6 The set of linear mappings 8 of R into R such that

9McL,

8 L = (0)

is denoted by 0.It is a linear Lie algebra. Because O2= {0},one gets [0,O] = (0). Clearly

50.

259

CONDUCIBILITY THEOREM, E. E. LEVI’S THEOREM

50.6.1

f(u)

-

g(u) E 0

for u E G.

The equivalence offand g will be settled by some 1 - K , with K (1 - K (Note that (1 - K ) ( I 50.6.2

+

K) =

) f ( M

E

0 such that

+ 4 = g(4.

1.) This condition on

[f(u),K ] = g(u) -f(u)

K

can also be written

for all u E C.

50.7 In 0 a new linear representationrof C i s defined by

A 4 9. = [f(4791

(=

k(4,791).

Then, because of 50.6.1, one can write 50.4.5 as

f(4

50.7.2

=

K

s(4 - f ( d

With 50.7.1 the desired result in 50.7.2 is nearly attained, but one still has to remove zf from the last member of 50.7.1. This will be done by solving 50.7.3

with respect to K

,f E

K

= ZCf.81

0. Note that according to 50.3.3 0 = 0+

a direct sum, where 50.7.4 50.7.5

Substituting z 50.7.6

f(4@ = {O}, f(u)

c

zf@= {O},

zfnondegenerate on 0,.

~from ~ 50.7.3 * into ~ ~50.7.1 and observing (50.3.1)

r(4zf

=Z f m ,

one gets as an equivalent condition on K

260 50.7.8

39-50.

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

f ( a ) K - (g(a)-f(a))

E

@

According to 50.7.4, f ( a )9. = SEf(a)

for all a E G. for 9 E @.

Because of their reducibility of fL andf,, this means that 6 = 0 or 6 causes an (essentially unique) equivalence offM andf,, and dim @ = 1. In the first case @ = {0),zfis nondegenerate and 50.7.3, and consequently 50.7.2, can be solved. Tn the second case 50.7.3 can still be solved mod @. Because of 50.7.4,this solution fulfills 50.7.7exactly. This gives a K such that

3(4

K

= g(a) -f(a)

+ ~ ( aa,)

where 0 # 8, E @ and the scalar u(a) depends linearly on a E G. Computing A'([% 61) K = S ( 4 m K -3(4f(4K , while taking 50.6.1 into account, one gets for the second member [f(a),g(b)-f(b)l - [f(b),g(a)-f(a>l = [g(a),g(b) [f(b),g(a)-fWl = g([a,bl) - f ( [ a ,

-m1b1)9

and for the first member, g([a,4)-f([a,

4) + m, bl) 9.0.

This shows that u vanishes on the commutator algebra of G, which because of the semisimplicity is G itself. Therefore 50.7.2is solvable, which verifies the assertion of 50.5 and therefore of Theorem 50.1. 50.8 ToproveTheorem 50.2 oneconsiders G EAlg Lie Lin Com presented in R E Spa Lin Com with an invariant L sub R . Again there is an M sub R and a linear representation g of G in R such that 50.8.1 50.8.2

L+M=R, g ( a ) M c M,

L n M={O},

g(a)x = ax

(a - g(a)) R c L

for x E L ,

for a E G ;

g is unique up to equivalence. Proposition Under these conditions, suppose that g ( C ) is semisimple and irreducible on both L and M. Then M and g can be modified such that g ( G ) sub G.

50.

CONDUCIBILITYTHEOREM, E. E. LEVI’S THEOREM

26 1

This will be proved in 50.9-10. 50.9 The set of linear mappings 9 of R into R such that

9McL,

BL= (0)

is again denoted by 0. Again [@, 01 =

w,

and a - g(a) E

0

for all a E G.

The kernel of g is denoted by A . If a E A , then g(a) = 0; thus a E 0. Hence A

= 0 and A abelian.

50.10 Take a linear subspace E of G such that G= E + A ,

EnA

= (0).

In this splitting the E-component of a E G is denoted by e(a). Thus, e(a) = a mod A ,

e(a) E E,

a - e(a) E

0.

Clearly e need not be a representation of G, but 50.10.1

4 [ a ,bl) = [W,e(4l + #(a,b)

where #(a,b) E A .

The desired modification of g will be obtained by replacing g(a) by g(a) + [g(a),w]

with a suitable w

E

0.

This does not influence the second and third relations in 50.8.2. Since Ma) + [g(a),wl, db) + [g(b)9wlI = [g(a),&?@)I + [g(a), [g(b),wl1 - [g(b),Ma), all = k ( 4 , db)l + “g(a), g(b)l,wl,

the new g(G) will again be a Lie algebra. The only thing one has to arrange is that 50.10.2

g(a>+ [g(a), wl

E

G.

Then the new M is obtained as (1 - w ) M , which is indeed invariant under the g ( 4 + [g(a),wl.

39-50.

262

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

To solve 50.10.2 one applies 50.4.4, with g,e instead off,g. Then for c E G,

C ([[g(c>,g(a’>I, e(aj>l+ [g(a’>,e([c, ajI)I>= 0 J

which by 50.10.1 becomes

C (“g(~),g(a’)I,e(aj)I + Ma’), I

[e(c),e(aj>l+ ~ ( caj)l) , = 0.

The Jacobi associativity law applied to the first summand leads to [g(c)iz [ ~ * ~+ ’ I C ( k ( 4 , [e(aj>,g(C>- e(c)ll +

U(C, aj>l)= 0.

I

Again =

C [g(a’>, [g(a,),$Il I

=

CI [&?(a’>, [e(a,),911

defines a linear mapping za of 0 into 0. By the same argument as in 50.7.4-5 0 splits directly into @, where zg is nilpotent, and into @, where za is nondegenerate; both are g(a)-invariant. Similarly, g(a) @ = (0) and dim @ G 1. Again zi = ZCe.el

can be solved by

K

mod @

E za 0 and Za K j = U(C, aI)

by

mod @

K, E za 0 n A because A is za-invariant. Since the g(b) commute with the elements of @, this leads to

[g(c>,z“.] + zB(g(c>- 44) + C ([g(a’),z”,I)

=

I

0

and, since za commutes with g’(b), to z~([s(c>,KI + g(c) - e(c>+ C Ig(a’),~jI)= 0. I

Now e(c> - C Ma’), I

~ j l =e(c>- C [e(a’),~jlE G ; I

hence g(c>+ [g(c),.I

+ 4g(c)P, E G

for some linear function u and some fixed 9, E @. The commutator argument again shows that u = 0, which proves that g(c)

+ [g(C),K] E G

for all c

EG

and thus proves 50.10.2. This completes the proof of Proposition 50.8.

50.

CONDUCIBlLITY THEOREM, E. E. LEVI’S THEOREM

263

50.11 Proposition 50.8 is the basis of an induction by which Theorem 50.2 will be proved. At any step in the sequel the existence of the Levi-splitting for lower dimensional Lie algebras will be taken for granted. Let G E Alg Lie Com and let A be the radical of G. Let B be an ideal of C within A and B # {0},# A . Then G mod Badmits a Levi-splitting; thus

G = F, + A ,

where F , n A

=B

c F,,and F , mod B is semisimple.

Again F, admits a Levi-splitting. F , = F,

+ B,

where F, n B = (0) and F, is semisimple.

Thus G = F, + A and a Levi-splitting of G is obtained. As a consequence, it may now be supposed that the radical A of G does not contain an ideal of G different from A and (0). Thus A is abelian and acts irreducibly on A . Now suppose that Gmod A is not simple. Then there are proper ideals GI,G, of G, different from A , such that 50.11.1

G = GI + G,,

G , n G, = A .

There are Levi-splittings 50.11.2

G, = F,+ A with semisimple F,.

YYEF, YrEc3xleaves G, and A c G , invariant, hence because of its conducibility, also a linear subspace K of G, such that 50.11.3

G,=K+A,

KnA={O}.

Because of [F,,F,] c A , one even gets

[FI,KI

= {Oh

K generates a Lie algebra K ‘ , which still fulfills

50.11.5

K c K ’ c G, = K + A , K nA

= {0},

thus there is a linear subspace A’ of A such that 50.11.6

K ‘ = K + A’.

39-50.

264

ALGEBRAIC APPROACH TO LINEAR REPRESENTATIONS

Now [A’, A’] c A n K’ c A’. This shows that A’ is the radical of K‘.There is a Levi-splitting 50.11.7

K’ = L + A’ with semisimple L.

Now, because of 50.1 1.1-7, G = G,

+ C2= PI + G2= PI + K + A = Fl + K ’ + A = Fl + L -,

A

and, because of 50.1 1.4, 7, F = P,+ L semisimple and G=F+A, a Levi-splitting. Therefore G mod A may be assumed to be simple. Consequently G may be supposed irreducible on both A and G mod A. If G and G are isomorphic, Proposition 50.8 can be applied to G acting in G. The role of the “clean” representation is then played by the reduction mod A. Proposition 50.8 guarantees the existence of a subalgebra P isomorphic with G mod A. If, however, G has a nontrivial center, the center is in A and therefore equals A. Then Gis semisimple and the conducibility theorem applies to the effect that A possesses a G-invariant linear complement, which is even an ideal. This proves Theorem 50.2. 50.12. Historical Note H. B. Casimir and B. L. van der Waerden [Math. Ann. 111, 1-12 (1935)l proved the conducibility theorem with the use of the Casimir tool and weight theory. J. H. C. Whitehead [Proc. Cambridge Phil. SOC.32,229-237 (1936); Quart. J. Math. 8,220-237 (1937)l gave a more elementary proof in which no appeal was made to weight theory, and by the same method he proved E. E. Levi’s theorem. It seems that this was the first correct proof of this theorem. The present proofs, based on the same ideas as Whitehead’s, were first published by H. Freudenthal [J. Madras Univer. B27, 225-236 (1957)l. They could be simplified by the use of cohomology language.

51-62 REALITY IN LIE GROUPS AND ALGEBRAS AND TH E IR LI N EAR R E PR ESENTAT I0 NS

51. MAXIMALLY COMPACT DRESSING Of any complex semisimple Lie algebra only two real types have been explicitly dealt with, the unitary and the standard ones. The goal is now a complete classification of real types, as given in Section 26 for the complex case. It is expressed as a classification of involutory semimorphisms. Its theory is developed in the present section. 51.1-3 contain general remarks. From 51.4 onward G is supposed to be a complex semisimple Lie algebra with the involutory semimorphism C . Also, G is supposed to be linear, the infinitesimal algebra of a linear Lie group G ; however, i n assertions and proofs concerning C alone, this is only a matter of notational convenience.

51 .I-3. Preliminaries 51 .I. Proposition If @ is an automorphism or semimorphism of G E Alg Lie (G E Gru Lie Lin), then N

N

@a = @Z

(@a= @Z@ I ) ,

@-I

Remark The second statement can also be derived from the first, e.g. by using exp. This shows that the second statement also holds for local automorphisms and semimorphisms. Thus, even local automorphisms and semimorphisms normalize Int G. N

Proof

@ax = [@a,x]

=

@ [ a ,@ - I

x]

=

@Z 0 - I

x.

N

@ax = (@a)x(@u)-l

=

@(a(@-]x) u-,)

=

@Z @ - I x.

51.2 Any real Lie algebra can be obtained by real restriction from a complex one, namely, from its complex extension. Two isomorphic real Lie algebras GI,G2 possess isomorphic complex extensions. Then the question is, when do two involutory semimorphisms C,, C, of complex Glead to isomorphic real C , , G 2 ?Such an isomorphism of GI onto G, extends to G a s an automorphism A of G with the special feature that x = C,x implies A x = C, A x , hence AC, = C, A , first on G I ,then on G. This suggests the following definition. 265

266

51-62.

REALITY IN LIE GROUPS A N D ALGEBRAS

Definition The involutory semimorphisms C1,C2of G E Alg Lie Com are called isotypic if there is an automorphism A of G such that

C2 = ACl A-’,

and inner isotypic if A can be chosen as an inner automorphism. Proposition Two involutory semimorphisms of G E Alg Lie Com lead to isomorphic real restrictions iff they are isotypic. “Only if” has been proved; the proof of “if” is the same argument reversed. Classification of the real types of G up to isomorphism comes to the same thing as that of the semimorphisms of G up to isotypism.

51.3 If G E Alg Lie Com SS has the semimorphic image G’ by means of C, then G and G’ are isomorphic, for example by means of CC,,, where C,,, is some unitary semimorphism of G.

The equivalence of the following conditions was proved in 1.12. Definition I: E Alg Lie Rea SSS is described as a twin type if it derives from G E Alg Lie Com S S S by waiving, or, equivalently, if its complex extension is not simple, or, equivalently, if it is the C-restriction of a direct sum G + CG, where CG is a copy of G E Alg Lie Com SSS. If L E Alg Lie Com, the result of waiving in L is denoted by L**. Proposition F E Alg Lie Rea SS is a direct sum of twin-type algebras and algebras with a simple complex extension.

Proof P is the C-restriction of some G E Alg Lie Com SS, which splits into simple direct summands G,. The splitting is C-invariant. Summands G, # CG, account for a twin summand of F,whereas G, = CG, produces a summand with simple complex extension. 51.4. The Class of CC,, Proposition Let C be an involutory semimorphism of G E Alg Lie Com SS, and C,, a unitary one. Then (CCu,)2E Int G.

Proof 33.9(6) and 33.3.1 grant the existence of an A E Aut G such that CC,, E A * Int G, and AC,, = C,,A. Thus, using Proposition 51.1,

C,, C = C,,(CC,,) C,, E C,, A * Int G * C,, 1 = C,, C * CC,, E A2.1nt G,

= C,,

hence, A 2 E Int G, which proves the assertion.

AC,, Int G = A Int G,

5 1.

267

MAXIMALLY COMPACT DRESSING

51.5.1-10. C-Third Dressing and the Hermitean-Unitary Split

In further investigations it proves convenient to have C commute with C,,. 51.5.1. Definition For an involutory semimorphism C with C H = H a third dressing with respect to H i s called a C-thirddressing if the C,, belonging to the dressing fulfills

cc,,

=

c,, c.

The existence of such a dressing will be shown in 51.6. Meanwhile a few properties of a C-third dressing are derived. First a few definitions elaborating on the notions of Section 38. 51.5.2. Definition u E Gunt)Cu,u = u,

s E Gh,

C,,s

t)

= -s

-

s E Sun.

Gun infinitesimally generated by C,,,

Ghe = exp Ghe.

The elements of Gheand Gheare called hermitean. For a C,,-invariant trunk H (cf. 33.13), Hhe =

H,, = H n

Hhe= H,,

=H

ch,,

n Ghe infinitesimally generated by Hhe.

51.5.4 For an involutory semimorphism C and a C-invariant trunk H : Definition Gc = toeC(Cu= a),

G,

infinitesimallygenerated by Gc.

Hc = tosdCh = h), Hc infinitesimallygenerated by Hc.

Note that C need not be extendible to an automorphism of (real) G, though it is if, for example, G is simply connected or centerfree. Even if C is extendible to G, then still G, is only the 1-component of the group of C-invariant elements of G . 51.5.5. Proposition If CC,,

= C,,C,

51.5.6. Proposition If CC,,

= C,,

CHh,

= Hhe.

then CG,,

= Gun,c c h , =

the.

C , CH = C,,H = H, then CH,,= H,,,

268

51-62.

REALITY I N LIE GROUPS AND ALGEBRAS

51 3 . 7 . Definition If CC,, G C . u n = GC

= C,,C,

Gun,

then set GC.he = G C

Gc,Uninfinitesimally generated by G,,,,,

the, Gc,he = exp GC,he.

Furthermore, if C H = C,, H = H, then set H c , u n = HCn

Hun, HC,he = HC Hhe, H,,,, infinitesimally generated by HcVun, &he

infinitesimally generated by HCehe.

Finally, Hc,,n,comandHc,h,,,o,willdenote thecomplexifications of Hc,,,and HC,he in H respectively, and HZ,,, H&n,com,H2he,H ~ h e , Cthe o mI;-images r of H c , u n , Hc.un.com, HC.her HC.he.Com in H*, respectively. 51 5 8 . Proposition Suppose that CC,,

=

C,,C. Then

G = G,, + G,, direct as linear spaces, (2) Gc = G,,,, GC.he direct as linear spaces, (3) Gun= Gc,,, + iGc,hedirect as linear spaces, (4) #C(GC,uni GC.he) = O.

(1)

+

Proof

(1) was proved in Section 38. (2) The splitting u = s + u (s E Ghe,

C,,)is unique; Cs E Ghe,Cu E C,,(see 51.5.5), Cu = Cs + Cu, therefore, if Cu = u, then Cs = s, Cu = u. (3) u E G,, is the sum of +(1 + C ) u E C,,,,and

uE

-$(l

+ C)iu E iCc,,,; (4) follows from 51.5.3.

Remark C = C,,iff GC,he = (0).

51.5.9 As soon as the existence of C-third dressings has been ascertained, Proposition 51.5.8(1-2) may be stated as follows: Theorem Up to isomorphism the real restrictions Gc of C

E Alg Lie Com SS arise from G,, by splitting G,, directly into subspaces K , L such that

[K,KI c K, and putting Gc = K

[K,LI = L,

[ L ,LI = K,

+ iL.

Guninto K = C,,,,, L = iGC,herand verify the above relations. Conversely, given the splitting with the above properties, one can easily show that [ K + iL,K+ iL] c K + iL, hence that K + iL is a real restriction of G.

Proof Given G,, split

51.5.10. Theorem As a n analytic manifold, C, is the product of its closed submanifolds Gc,,, and G C , h e by means of the multiplication in C; GC,he= Gc n Ghe, and CC,,,,= Gc n Gun.

5 1. MAXIMALLY COMPACT DRESSING

269

The proof runs along the same lines a s that of Theorem 38.4. Note that G, is closed, as shown i n 38.5(4). 51.5.11-1 2. The Maximal Compact Subgroup 51 -5.11. Theorem G,,,,, is a maximal compact subgroup of G,. Proof The compactness is evident. According to Theorem 51.5.10, any subgroup of G, larger than G,,,, would contain an element # 1 of Ghe,which cannot generate a relatively compact subgroup. Remarks ( I ) In fact, every maximal compact subgroup of G, can be obtained i n this way; all are conjugate within G,. This, however, is a rather profound fact, which will be proved in 65.4. (2) If, instead of G,, one considers a group 6, which is a wrapping of G,, then corresponding to G,.,, (generated by G,,,,) may cease to the subgroup be compact. I n fact, if 6, is the universal wrapping of G,, then 6,,,, is some wrapping of G,,,, (actunlly, it is the universal one; see 62.3); it may happen that C,,,,, is not semisimple but has an abelian direct factor; then 6,-,,,, is not compact.

c,,,,,

51.5.12. Theorem C,,,,, is its own normalizer in G,. Proof so normalizing G,,,, may be assumed i n GCShe,hence of the form so = exp so with some so E CCShe. For u E G,.,,: sousot = u I E G,,,,. Thus u-IuI so = ii-Is0u E GC,he.Because of the uniqueness of the heriiiitean-unitary split, this shows u-Isou = s o for all id E G,.,,,; hence [u,s,] = 0 for a11 u E G,,,,,.

In #(.?ou,s)

+ #(u,3,s)

=0

the first summand vanishes for all u E G , ~ , ,and all s E G h eTherefore . Fos belongs to the orthoplement G,-he of G,,,,,. On the other hand, i0s E G,,,,,. Hence ios= 0 apart from i o u = 0, which shows that Yo = 0; hence so = 0, so= 1. 51.6. C-Third Redressing Definition If C H : H , then define C * by

(c*()(/I) =

m,

for all /I

E

H,

8 E H*.

Clearly C* is linear and belongs to Aut W * . If H is a C-invariant trunk and C,, is based on H , then C and C,, commute if restricted to H , since C permutes W* and consequently leaves HE,and H,, invariant.

51-62.

270

REALITY IN LIE GROUPS AND ALGEBRAS

By a suitable choice more can be attained. Theorem A given third dressing on a trunk H

=

CH can be changed into a

C-third dressing on the same trunk. Proof A provisional C,,, based on H has to be changed into gC,, some g = exp g, g E H,,, such that

g-' with

cgc,, 2-1 = gc,,2-1 c, equivalently, 51.6.1

(pc,,C)2=

With some scalars K,, Ce,

= K,

1.

+,.

From C2= 1 it follows that 51.6.2

from [e,,e-,]

= h,,

51 -6.3

from [e,,ep]

=

N,,pe,+B and INa,Sl = iNc*a,c*pl (cf. 23.21,

51.6.4

K,

' KB

I?p = K a + p t?,+p.

This shows the existence of h E If,,with exp a(h) = K , i z

for all a

E

W*.

From 51.6.2 it follows that 51.6.5

exp(a

+ C* ct)(h)

=

1;

thus, because of h E Hst, 51.6.6

(a

+ C* a)(h)= 0.

To satisfy 51.6.1 its first member with g = exp g is applied to e,, which gives kc,,I?, exp 2(a - C* a)(g). e, = K , I?, exp 2(a - C* a)(g).e,

because of 51.6.2-3. To fulfill 51.6.1 one must solveg from 51.6.7

2(C*

-

a)@) = a(h).

For this one need only take -ah for g, as follows immediately from 51.6.6.

5 1. MAXIMALLY COMPACT

27 1

DRESSING

51.7. Kinds of Trunks The conjugacy theorem for trunks is a particular feature of unitary and complex semisimple Lie algebras (thus also of the twin type). In all other cases different kinds of trunks have to be distinguished. A first orientation of the variety of possibilities is furnished by the following Proposition Let the trunk H be C-invariant. Then up to conjugacy within G, every C-invariant trunk H'is of the form QH,where u belongs to the unitary restriction determined by a C-third dressing on H. Remark Of course, not every iiH provides a C-invariant trunk. Proof With C-third dressings on H a n d H', one obtains C,, and C;,, both commuting with C. By 33.13 they are conjugate, N

C:,

=

EC,,a"-'

= &(C,, a - l )

C,,,

such that c?H = H'. N

Now E(Cu,a-') also commutes with C. With the hermitean-unitary splitting a = su,

C,, s = s-l,

C,, u = u,

one finds CS2= S2 C . If s = exp s for some (unique) s E Ghe, then, by the Cinvariance of Ghe,Cs = s also, whence s E GC,hc by 51.5.10. Now F ' H ' = QH is a C-invariant trunk of the kind wanted. 51.8. Maximally Compact Trunks H , of 'a C-invariant trunk H splits called, redirectly into a torus group H,,,, and a flat abelian group spectively, the torus part of H , and the flat, hermitean, or standard part of H , (and correspondingly for H,). (See 30.5.) Note that the flat part would not be uniquely determined if H , were considered as a real group, disregarding the complex structure ofthe H i n which it is embedded; in the present case, however, it is uniquely determined. Definition A C-invariant trunk H of G is called maximally (minimally) compact (with respect to C) if H , contains a maximal torus of G, (if no proper subset of the torus of H , can be the torus of H i for another C-invariant trunk H'). Maximally compact trunks are studied here, minimally compact ones in Section 60. The existence of maximally compact trunks is granted by the following Proposition Every torus of G, is in a maximally compact trunk.

T into a trunk H of G. Take h regular in H. Then Ch is also regular. Among the ~h TCh is a Proof Take a maximal torus T in C,. Then 35.9 embeds

+

272

51-62.

REALITY IN L I E GROUPS AND ALGEBRAS

regular one, say, h‘. Now h’ is C-invariant and determines a C-invariant trunk H ‘ ; hcommutes with every element of Tand so does Ch, since T is C-invariant. Therefore h’ commutes with every element of T,and consequently T = H‘ = C H ‘ . Since T was taken as a maximal torus of G,, H’ is a maximally compact trunk. 51.9. Criteria on Maximally Compact Trunks Theorem The following assertions on G E Alg Lie Corn S S in third dressing and the semimorphism C are equivalent: (1) H i s a maximally compact trunk (with respect to C).

(2) (3) (4) (5)

No a E W* vanishes identically on the torus part H,,,, of H,. The torus part H,,,, of H , contains a regular element of G. CC,,a # -a if a E W* and if C,, is defined by a C-third dressing on H . CC,, (as in 4) leaves some chamber invariant.

Proof One may suppose that CC,, = C,, C anyhow (51 -6). 1 2 + 1 4 (with the same a ) : Suppose a(h) = 0 for h E H, n Hun. Then a(h CC,, h) = 0 for all h E H,, and, since a is linear, for all h E H . Thus ((1 CCun)a)(h) = 0 for all R E H which falsifies 4 with the same a. 12 + 1 1 (using 14): Put B = H,.,,,. Suppose that a(B) = (0) for some a E W*.Then [B,et,] = (0). From 14 it follows that C * a = -C;,a, hence Ce, = v,e, with Iv,I = 1. Putting v;l2e, = u, one gets Cu = u. Putting b = u + Cunu,one gets b = C,,b = Cb # 0, and [B,b]= (0). Now exp 76 (T real) E G,,,, and commutes with every element of B. Therefore B and exp ~b (real T )

+ +

generate a torus, which proves the existence of a torus subgroup of G, larger than B. 2 --f 3 is trivial. 3 + 1 : Let h E H,.,, be regular and let B c G,, 2 H,,,,, and B maximal compact abelian. Then [h,B]= {0), and, since h is regular, B c H . Since B is compact, B c Hun n G, = H,,,,,. Thus H,.,, is maximal compact abelian within G,, which proves that H is a maximally compact trunk. 4 + 5 : h’ = (1 + CC,,)h for R E H,, is CC,,-invariant; a(h’)= ((1 + CC,,)a)(h).Therefore h can be chosen such that a(h’) = 0 for no a E W * . lh’ is in some chamber, which is also CC,,-invariant. 5 + 4 is obvious. Then 1 4 + 12: one gets

Suppose that CC,,a

= -a

for some a

E

W * . For h E H,,,,

a(h)= a(CC,, h) = (CC,, a)(h)= -a(h);

therefore a(h) = 0.

51.

MAXIMALLY COMPACT DRESSING

273

51 .lo. Conjugacy of Maximally Compact Trunks Theorem Two maximally compact trunks of G (with respect to C) are coiijugate by means of elements of G,.

C-third dressing with respect to the trunk H is supposed. 51.7 allows one to restrict the proof to the trunks H and H ‘ = 11H = CGH, where 14 E Gun.Let B be the torus part of H, and B’ that of HL. Then

Proof

BcG,,,nG,=F; G-IB’ is compact and contained in H, hence in H,, c G,,,; thus i i - l B‘ c Gun,

B’ c C,,, as well as B’ c C,. Therefore B,B‘ are both in F. Now F is the infinitesimal algebra of the compact linear Lie group F = G,,,,,, which according to 19.15 splits directly: F=Fo+F, with abeliun F, and semisimple F , . Furthermore, B,B‘, as maximal abelian subclgebras of F, split directly:

B = Fo

+ B,,

B ‘ = F,+ B ; ,

where

BI=Fl n B,

Bi=FI n B ’ ;

B, and B ; are necessarily trunks of F. By virtue of the conjugacy theorem in F , , there is an a E F c G, such that B; = GB, and consequently B’ = GB.Now both GHand H’ are trunks containing B‘, which possesses a regular element: hence GH = H ’ as desired.

51 .I1. Maximally Compact Dressing Definition

A C-third dressing on C H = H is called a maxiinally compact

dressi/ig if H is maximally compact. The term ordered m a s i m 7 / / ~coriipact

tlressirig is reserved for a choice of the partial order which makes a CC,,-invariant chamber of H: dominant. In 51.6 and 51.8-9 the possibility of ordered maximally compact dressing was proved. Theorem

51 .I1 .I

Under ordered maximally compact dressing on H

C = I;AC,,,

where A is the straight extension of an automorphism of W + +(see 33.3.1), 11 E H,,,,

51-62.

274

REALITY IN LIE GROUPS AND ALGEBRAS

51.11.2

A2=1,

51.11.4

Ah = & - I .

N

Proof CC,, leaves W + +invariant and induces an automorphism A of W + + with A2 = 1, according to 51.4. A extends according to 33.3.1 to C, and to H;

CC,, A leaves H elementwise invariant and preserves Gun.By 33.9(2) it is inN

duced by some h-' E Hun.Finally 1 = C 2= hAC,,KAC,,

= KAh.

51 -12-1 5. Reductions toward Classification 51.12 With a view to isotypism the presentation of C according to Theorem 51.11 can still be simplified. This will be done in 51.12-14. Proposition Up to inner isotypism of

C = KAC,,, as presented in 51.1 1, h = exp h,h E Hunmay be supposed to fulfill

P W =0

if p # Ap, if p = Ap,

p(h) =0,7ri

for p

E

W++;hence Ah = h.

Proof In the given C = KAC,,, because of 5 1.11.4,

h = exp h, 51.12.1

p(h

+ Ah)

=

h E Hun,

0 mod 27ri

for p

By means of ho = exp ho,

ho E H u n ,

C should be replaced by inner isotypic N

C'= K O CK,'

=

KO Ah;'

C.

Putting exp h'

= h' = h ho(Ahil),

E

W++.

5 1.

MAXIMALLY COMPACT DRESSING

275

one gets the new C' in the form N

C' = h' AC,,. As to h', one can take it as h'

=

h

+ ho - Ah,;

ho is chosen such that for each two elements of W + +interchanged by A , p(ho) = -p(h)

for one element p among them

=0

P (A,)

for the other one.

Then with a view to 51.12.1 p(h')

=0

mod 7ri

p(h') = 0 mod 27ri

for Ap = p, for Ap# p ,

Without changing h', one can modify h' such that p(h') = O

or

7ri

p(h') = 0

for A p = p , for Ap # p .

This proves the assertion. 51.13 If A

=

1, the reduction of

c = LC,,

N

= exp h

C,,

can be continued by the use of some S E Int W*, which extends to some ii with u E Gun;C is replaced by inner isotypic N

C' = iICD-'

= iihii-'

C,,

= exp

Sh C,,,

For h this means that it may be submitted to the action of Int W* as well as replaced by h' such that p(h) = p(h') mod 27ri for p E W++. In the terminology of 33.14.4 this says the following.

Proposition In c = LC,, (h = exp h, h E H,,) up to inner isotypy of c one can change h into any Int mod W*-equivalent. In C = LAC,, (see Proposition 51.12) the same change is allowed with respect to the subgraph of A-invariant elements of W++. 51.14 Thus h may be assumed in the closure of the principal domain b (or in that belonging to the A-invariant part of the graph). Whence:

276

51-62.

REALITY I N LIE GROUPS A N D ALGEBRAS

Theorem After ordered maximally compact dressing of C E Alg Lie Com SS with respect to the involutory semimorphism C, up to inner isotypy by nieans of an element of Gun,C can be assumed to be such that

51 .I 4.1

C = KAC,,,

where A is the straight extension of an automorphism of W + ' ,

51 .I 4.2

AW"

51 .I 4.4

A'=

51.14.5

h = exp 2 ~ i h ,

51 .I 4.6

p(h) = O

51 .I 4.7

p(h)=O,t

51 .I 4.8

a(h) G 1

CY",

=

for p

1,

h E Hher E

W++, p #Ap,

for P E W + + , p = A p , for every

o! E

W' which is a sum of

A-invariant elements of W' '.

51.15 Further simplifications are possible for simple G. A=l Proposition If, under the conditions of 51.14, C is simple, 51.14.8 can be replaced by

51 .I 5.1

&(h)Q I

for the top rootform G.

This is a consequence of 25.6. Table E of top rootforms shows that to satisfy 51.1 5.1 p(h) = is allowed for two primitive rootforms p at most and that such a pair of nonvanishing p(h) is only possible for

+

A,,

D,( P I 9 P 2 . P A

E6 ( P l 9 P 3 ) .

However, even in these cases, the number of nonvanishing p(h) will be reduced to 1 by the use of Proposition 51.13; then

51 .I 5.2

p(h) # 0

for at most one p

E

W++.

A,: Suppose that i <j , pi(B) = pj(h)= 4, pk(h) = 0 for k # i , j . Put a =

51. pi

MAXIMALLY COMPACT DRESSING

+ p i + l+ . . . + pj-l. Thus a

E

W+.Applying S,, one gets

PdSa h) = (Sa PA (h) = (pi-, a)@) =

+

+

for k = i - 1 for k = i for k = j - l # i

= (pi - a)(h) = 0

=(pj-,-n)(h)=-) = (a =

277

+ p,)(h)

=

for k = j

1

0 otherwise.

The unit appearing for k = j can be replaced by 0 without changing h. By applying S, the pair of primitive roots for which h does not vanish moves to the left. Repeating this procedure one finally arrives at the announced result. D,: Let p(h), u(h) # 0 for some distinct p,u E W++.Then both are endpoints of the graph, and at least one of them, say p , is the endpoint of a short branch. Let ct be the sum of the elements of W++different from p, and let T be the third endpoint. Then Sap = p a, S,u = u - a, S,T = T - a. Thus

+

p(S,h)

= u(S,h) = 0

mod 1,

+

T ( S , ~= ) mod I ,

the other primitive rootforms being 0 in S,h. So application of S, yields the required result. E,: Here the only troublesome case is with p l ( h ) = p3(h)= 3. Then one can for instance apply consecutively

A f l

In any event, the subgraph of A-invariant elements of W + +is of the kind A. By 51.14.6-8, p(h) # 0 only happens for a p of the subgraph and then at most twice. One can again attain 5 1.15.2 by the same procedure as in the case A = 1. A as required in 51.14.2-4 is uniquely defined except for D,. Even in this case, up to outer automorphisms, it may be supposed to be the plus-automorphism of33.11 (interchanging p I , p 2 in D4).Then admissible h, according to 51.14.5-7, 51.15.1-2, may be supposed to fulfill A,(I=.l): p j ( h ) = O pj(h) = 0, D, :

+

forall j # + ( I + l), for j = $(/ $- I), if I is odd.

.P l(h) = p

m = 0, pj(h) = 0, f for j # 1,2,

yet at most one 3 only.

51-62.

278

REALITY IN LIE GROUPS AND ALGEBRAS

p,(h) = O

E,:

for j Z 2 ,

p,(h) =O,-) for j = 2. (By use of the kaleidoscope group this comprises the case p,(h) Pdh) = 3.)

=0

for j # 6 ,

51.16. The Twin Case The twin case is conveniently fitted into the same frame. Let F E Alg Lie Rea SSS of twin type; then its complex extension FComis the direct sum G + CG with simple G and the involutory semimorphism C. If H is an ordered trunk of G, then CH, with transfer of the order, is one of CG. The ordered third dressing of G is mapped into one of CG by C. Definition In ordered third dressing of G + CG the choice of the trunk is H + CH and C* is assumed to interchange H: and CH; in orderly fashion (hence to interchange the dominant chambers); if the partial order is extended to a total one, H: is assumed to precede CH:. The third dressing is C-invariant. The plus-automorphism of FComis defined to interchange G and CG in an obvious way, that is by P,, = CC,, M , where M is the minus-automorphism. Then again C = AC,, with A = P,wM ;

F appears as the C-restriction of G + CG. 51.17. Classification Dressing Definition L E Alg Lie Com SSS in ordered third dressing. Let C be identical to L or to the direct sum of two copies of L (with the ordered third dressing defined as in 51.16). Let C,, be based on the trunk H of G. Let A be 1 or the plus-automorphismP (in the case G # L: P,,M). Let W++be enumerated as in 25.7 and 26.23. Let h = exp 27rih with p(h)

=0

for all p

E

for PI, p

E

W++(G)

or P,(4

=

3,

P(h) = 0

W++(G), p # p j ,

where ApJ = PJ

and p, occurs with a coefficient 9 2 in the top rootform of L.i t This conditioncould be omittedwhen defining the L,.It would not, however, lead to new classes.

51.

MAXIMALLY COMPACT DRESSING

If

279

c = hAC,,,

then the C-restriction of G is called of inner type if G = L and A = 1, outer type if G = L and A # 1 , twin type if G # L (and A # 1).

Inner types are indicated by L,, outer types are indicated by L,,*, twin types are indicated by L**, where j will be taken as 0 if p(h) = 0 for all p E W++. If L E L, then the class of real Lie algebras isomorphic with L,, L,,,,, L,, is denoted by L,, L,,*, L**. The class of real linear Lie groups with an infinitesimal algebra in L,, L,,*, L,, is called L,, L,,*, L**. The discussion led to the following theorem: Theorem Every G E Alg Lie Rea SSS belongs to at least one of the classes L,, L,,,, L**. It appears in ordered maximally compact dressing.

51.18-20. Isomorphisms 51.18 Different subscripts j with the same L may indicate isomorphic Lie algebras. The following is a method of finding such isomorphisms. Suppose a graph W + +with simple bonds only. Then by 33.2.4 all elements of W* are equivalent under Int W*. Let the top rootform d be a fundamental weight and let p, E W + +be such that ( $ , p i ) # 0. Let p j be some endpoint of the graph and P k its neighbor. Then pk(h) = f, p(h) = o mod 1 for p E W++, p # p k ,

is solved by (p,,p,)-Ihp,; this, under Int W*, is equivalent to h’ = (6,&)-‘/I; Since 6 is a fundamental weight, p,(h’) = 3,

p(h’) = 0 mod 1

for p

E

*

W++, p # p,;

hence L, = L,.

51 .I9 Still a consequence of 51,511 : Theorem If G

c = c,,.

E

Alg Lie Com SS and Gc is compact, then up to isotypy

51-62.

280

REALITY IN LIE GROUPS A N D ALGEBRAS

The compact ones in the classification of 51.17 are characterized by the lack of an asterisk and the subscript 0. 51.20 In the detailed classification that follows in Sections 52-53 the various real C, EAlg Lie Rea SSS will be described by specifying their G,.,,, that is, the infinitesimal algebra of a maximalcompact subgroup G,,,, of G,. The structure of G,.,, can be used as a criterion of nonisomorphy of the indicated types, though in almost all cases simpler criteria like the signature of the restricted Killing form are available.

of CC,, in C,,, = -1-eigenspace of CC,, in G,,, signature Gc = -tr CC,, = dim GC,he- dim C,,,, C,,,,

= +I-eigenspace

= dim

C, - 2 dim CC,,".

51.21. The Description of the Graph of C,,,, C,,,,always will be either semisimple or the direct sum of a semisimple algebra and a 1 -dimensional abelian algebra, mutually orthogonal under I,&. The intersection of H with the (interpreted within C ) is a trunk, and the dual of semisimple part of Gc,un,com this trunk is interpreted within c H * by means of 5. In particular, the rootforms of Gc,unfor the rootforms of the semisimple part of Gc,un,com, short, are now elements of H ~ ~ u n and ~ c oeven m of iH&,. The restriction of $G to Gc,un.com is invariant under the adjoint of G,,,,, and is simple semisimple (and consequently ~ c , u n , C o m is therefore if Gc,un,com irreducible) ~ , ! J ~ equals J ~ ~ #Cc,ua,com , ~ ~ , up ~ ~ to ~a constant factor # 0. In general, the same is still true on every simple semisimple direct summand of CC,un,Comr whereas the summands themselves are orthogonal to each other according to both quadratic forms. This property, if restricted to the trunk and transferred to the dual trunk by means of 5, leads to the following Proposition The inner product in the graph of C,,,, is, up to factors constant in every component, the restriction of the inner product in H*.

51.22. Further Notations and Conventions C,,,, will often split into two or even more summands. If they belong to the compact types of L, M, .. ., the isomorphism class of the sum is indicated by L + M + * To account for one-dimensional summands, one makes the convention :

-

a .

D, = the class of 1-dimensional complex Lie algebras.

This convention supplements the known equalities (see 25.8 and Table G): A1 = 81 = C , ,

6 2 = C2,

D2 = A1

+A,,

D3 = A,.

52. CLASSIFICATIONOF I N N E RTYPES

28 1

Any A,, B,, C,, Do,if it occurs, means 0 = the class of the null-algebras.

Cc,,,,causes a linear representation 6 in GC,he.In Sections 62-63 it will be shown that 9. determines the global structure of G,,,, and consequently the fundamental group of G,. For this later use 6 is indicated here by the top weight(s) of its canonical complexification. In this context a notation like 7Tk.L) + rrn(M)or n-k,rn(L + M) means that the first summand is represented ; r k and T,,, are the according to r k and the second according to r r nthen restrictions of the top weight of 6 to the trunks of L and M, respectively. It so happens that 6 is reducible iff G,,,,is not semisimple, in which case C,,,, contains just one summand of the class D,. Such a summand is always represented with two opposite weights, which fact will not explicitly be mentioned in the subscripts of T . A few more conventions:

51.23. Historical Note E. Cartan’s method of real classification [Ann. Ecole Norm. (3) 31,263-355 (1914) = Euures Z 1, 339-4911 was rather casual. F. Gantmakher [Mar. Sbornik 5 (47), 217-249 (1938)l built a theory of real classification on H. Weyl’s approach to semisimple Lie groups and Cartan’s further developments. His theory has here been refined and greatly simplified by the use of primitive roots and the notion of maximally compact dressing. The most elementary proof of 51.6 has been found by H. de Vries. The actual classification will now require a minimum of casual distinctions and computations. The notation 51.17, which shows great systematic advantages, was proposed and used long since by H. Freudenthal.

52. CLASSIFICATION OF INNER TYPES 52.1 G E Alg Lie Com S S S in ordered maximally compact dressing, G, of inner type. If G, E L, ( j # 0), then

C,,,,

=

real linear span of Hun and the T,e,

- ?,e-,,

where p, occurs in o!

51-62.

282

REALITY IN LIE GROUPS A N D ALGEBRAS

with an even coefficient,and

GC,hc= real linear span of the i(T,e,

- r,e-,),

where p, occurs in

o!

with an odd coefficient. The compact types Loare omitted.

52.2 A,,, (I > 1): The outer automorphism P (for I > 1) shows that A,,, and A,,,+,-, are identical. The same, however, can be established by an inner isotypism, which can be constructed by the method used in 51.15.2. The graph of G,,,, is

--

p1-pz-

-pJ-IY

*

+

f,+1-P,+2-

*

-p1.

+

C,,,, belongs to A,-I A,-, D,. Signature: 1 - (I + 1 - 2j)2. Top weights of (reducible) 6: d and -p,; that is ~IJ-,(A,-l

+ A,-, + D1)

and

~,-l,l(A,-l

+ A,-, + DI).

52.3 D,,, (I> 4 ) ; j # 0,1,2:A procedure like that of 51.15.2 shows that D,,, and D,,,+,-, coincide. Thereforej will be restricted: 3 G j G *(I+ 4). The graph of G,,,, is /fJ-l

P3-P4-

* * *

-PJ-2

\P,-l+ P,+1-P,+2-

-

* *

2P, + - * *

+ 2P,+

PI

+ P2

--PI

\P* F o r j = 3 the first component of the graph is nonexistent; f o r j = 4 it reduces to the last two dots; f o r j = 5 it reduces to the last three dots. C,,,, belongs to DIP, D,-J+2(for j = 3,4,5 because of the conventions made in 51.22). Signature: I - 2(1+ 4 - 2j)2. Top weight of 6 arising from p3 + * * + p , + 2p,+, + + 2p, + p , + p2 is

+

T3,3(D,-2 -k

'l-,+2)*

For j = 3, however, 6 becomes reducible with top weight

+

T~(DI D,-l) twice

arising from d and - p 3 . F o r j = 4,5, 7r3 must be interpreted according to the conventions 51.22. F o r j = 1,2,the types coincide by means of P.The isotypy can be established by an inner automorphism for odd I.

52. CLASSIFICATION OF INNER TYPES

283

D,,,: The graph of C,,,, is P3-P4-

+ + +

C,.,, belongs to At-, D,. Signature : -1. Top weights of 9: p3 2p4 772(At-,

* * *

-Pt-P2*

* * *

+ 2p, + p1 + p2, -p,

+ DI),

; that is

+ D1).

771-2(4-1

Note that for 1=4 the types Dl,l Dt,2,Dl,3coincide by outer automorphisms. 52.4 E6,,: Because of the coefficient 3 of p6 in d , j = 6 may be dropped. According to 51.18, j = 4,5 lead to the same type as j = 2, and, thanks to the plus-automorphism, j = 1,3 are the same, though this can also be established by an inner automorphism, using the method of 51.15.2. In addition to the compact type, j = 1,2 are left.

,

E6, : The graph of G,,,, is /p2

P3-PS-P6

*

P ‘4

C,,,,belongs to D, + D,. Signature: -14. Top weights of 9: d and -p, ;that is,

4 h + DI),

4 D , , + D,).

E6.2 : The graph of G,,,, is P3-PSP6-p4-pl, C,,,, belongs to A, Signature: 2.

+ A,.

Top weight of 9: d - p2;that is, T3,1(A5

+

52.5 E7,,: Because of the coefficients > 2 of pi in 6,j = 5,6,7 may be dropped. By 51.18, j = 4gives the same a s j = 1 . In addition toj=O, j = 1,2,3 are left.

E,, I : The graph of G,,,, is P2-P4-p6-P7

/” 3

‘PI

A

CX*

284

51-62. REALITY IN LIE GROUPS AND ALGEBRAS

Cc,,, belongs to D6 i Al. Signature: -5. Top weight of 9: d - p1 ;that is,

E7.2:

The graph of C,,,, is P3

I

P4-P6-P7-PS-P

1.

Cc,,, belongs to E6+ D,. Signature: -25. Top weights of 6:d, and -p2; that is,

nl(E6

+ Dl>i

T3(E6

+ DI>*

E 7 , 3: The graph of C,,,, is PI-PS-P7-p6-p4-p2-(p4

+ 2p6 + 3p7 + 2p5 + + 2p3).

C,,,, belongs to A , .

Signature: 7. Top weight of 9: p2

+ 2p4 + 3p6 + 3p, + pl + 2p, + p , ;that is,

52.6 E8,,: After an inspection of thecoefficients in the top rootform,j= 1 , 2 in addition to j = 0 are left. E8,I : The graph Of Cc,,n is P4

C,,,, belongs to E , Signature : -24.

+ A,.

Top weight of 9.: d - p ; that is,

b.2:

The graph of C,,,, is

(P3 + 2Ps

+ 3P7 + 4PE + 3P6 + 2P2 + 2P4k-PI-Pl-P5-P7-P8

/p6

‘P4

52. G,,,, belongs to D,. Signature: 8. Top weight of 8: pI

CLASSIFICATION OF INNER TYPES

285

+ 2p3 + 3 p , + 4p, + 5p8 + 3p, + 3 p , + p2; that is, TI(D8).

52.7 B l , j ( / > 3 ) : ForjfO,l,thegraphofG,~,,is /Pj-

p2-p3-

-pj-2

* ' '

'pj-, Pj-1 I-

' ' '

1

+ 2pj + . . . + 2p, + 2p,

* PI.

-PI

F o r j = 2 the first component is nonexistent; f o r j = 3 it reduces to the last two dots; for j = / the second component consists of the dot pI. C,,,,belongs to Dj-, + BIpj, with the usual conventions. Signature: I - 2(/ + 3 - 2j)(/ + 2 - 2j). Top weight of 19.: p 2 + . . . + p j + 2pj+l + . . . + 2p, + 2p, ;that is,

,

773.2(Dj-1 + B 1 - j + l > with the usual conventions, i n particular, for j 2TI(B,).

=

I, to interpret ~ ~ ( 6as, )

F o r j = 2, however, D becomes reducible with top weights Band - p z ; that is

+

r2(D, Bl-l) twice,

B l , l: The graph of C,,,, is / P I -t 2Pl p2-p3-

' * *

-pI-l

P'I

C,,,,, belongs to D,. Signature: /(3 21). Top weight of 8: p 2 + p 3 + . . . i-p l ; that is, ~

4%). Remark

B l + can l also be interpreted by admittingj = /

+ 1 in Bl,, and putting

6, = 0 . 52.8 C l , j ( / > 2 ) : Forj#O,I,thegraphofC,,,,is piPj+l-

. . . :pj-l ' ' '

-PI-I

0 (which in turn implies D 2= 1 for the chosen 0 ) . Conversely, still supposing K real, assume a provisional involutory semilinear Do on R such that 57.3.3

DOX = y ,

D ~ J=JX.

It exists, since x , y , if linearly dependent, have been assumed equal. Remember that

302 57.3.4

51-62.

REALITY IN LIE GROUPS AND ALGEBRAS

f ( u )x = y y for some y # 0.

Now g = Y (I Do f ( C 4 Do

is again a linear representation of G, which in the partial order transformed by C* has the top weight C*X equivalent to A. Thereforefand g are equivalent, and there is a linear mapping Tof R onto itself with

for all a E G.

T g(a) T-I = f ( a )

Since x is also a weight vector of g, one may even suppose that T-' maps x into a given multiple of x , say 57.3.5

Substituting the definition of g , one gets 57.3.6

TDOf(Ca)DOT-' = f ( a ) ,

and, substituting Ca instead of a, TDOf(a)Do 7'-' =f(Ca).

Combining these equations and putting 57.3.7

D = TDO,

one notes that D2 commutes with allf(a).

Thus, 57.3.8

D 2 = TDoTDo

is a scalar multiplier.

When applying 57.3.6 to TDoyy and replacing a by u, one gets TDOf(C4 w =fW TDO ??v.

Replacing yy in the first member according to 57.3.4 and 57.2.5, one gets KTDOX,

whereas in the second member, by 57.3.3, 57.3.5, and 57.3.4,

57.

303

SELF-CONTRAVALENCE

Hence EdlKI

TDox = lYl Y ,

dm TD0)’x d m TDO ( Y ( Y= d(3( y (TX =

=(K(

x

by 57.3.3, 57.3.5; hence by 57.3.7 D’x

= EX

and, thanks to 57.3.8, Dz = E ; 57.3.6 gives

@~(CU D-’ ) =f(a). If E = I , this shows thatfis virtually real. It also proves the other assertion of the theorem. 57.3.9. Remark It is easily seen from the irreducibility offthat D as asserted in the theorem is determined up to an arbitrary constant with absolute value 1. From the theorem and the first part of its proof it follows also that the reality of K and its sign are independent of the particular C-third dressing.

57.4. Theorem Every irreducible linear representationfof a standard type

C, and the self-contravalent irreducible representations of near standard or twin Gc are virtually real. Proof

By 56.1, if C = C,,, C*h=h;

but this is still true of near standard and twin C iffis self-contravalent because then by 56.2.2 and 56.1 C*h is equivalent to h and still dominant; thus C*X=h. To show that the irreducible representation with top weight h is virtually real, one considers its construction (44.5) with a view to reality. The semirnorphism C extends from G to the associative envelope b(G). But now the left-hand ideal annihilating the vector of top weight (denoted by -1 f M in 44.5) is C-invariant because of the following: (1) C* maps W +onto W + , and C interchanges the eOr(a E W’) up to real factors (kl). (2) c maps h - ~ ( hinto ) ~h - h(h)= ch - (c*A) ( ~ h=) ~h - A(c~). (3) C interchanges the expressions e!;P+l up to signs.

Therefore C extends to b(G) mod M as a semilinear involution which is compatible with C. This proves the theorem.

51-62. REALITY

304

IN LIE GROUPS A N D ALGEBRAS

57.5 Since the self-contravalence of an irreducible linear representation is preserved under isotypism and is not influenced by the choice of the trunk, one can apply the foregoing result to linear representations of any central or near central restriction, according to Proposition 54.4. Observe now that in ordered maximally compact dressing CC l eP =e-,,

Cnclep=e-pp

for p

E

Wi

',

and substitute these data into 57.2.1 and 57.2.5 for C = C,,and C = Cnc, : f(e,,

- - - eple-pk

* * *

x

= KX

and

f(ePpk . . ePple-pk '

* * *

e-,,J x

= K'

x

with positive K and K' because of the virtual reality of self-contravalentf. One can now forget about central and near central restrictions and state the following: Proposition If G E Alg Lie Com SS in ordered maximally compact dressing with respect to some C, iff is a self-contravalent irreducible linear representation of G, x a vector of top weight A, y a vector of weight C* A, and f(e-pu * * e-Jx is a multiple # 0 of y (py E W + + ) ,

-

then for inner Gc 57.5.1

and for outer C, 57.5.2

are positive multiples of x . Though proved only for simple G, the proposition applies in general. Twin factors do not contribute, and all simple factors behave independently, since their branches commute and every branch acts only on the component of the trunk to which it belongs. In any real simple factor where it occurs P is to be interpreted as the plus-automorphism. 57.6 To compute E for any self-contravalent irreducible linear representation fwith top weight h one must compare 57.5.1-2 with 57.6.1

f(Ce_,,

* *

- Ce-,,

-

cpr* * e-pl)x = K X .

Now in ordered maximally compact dressing (see 51.1 1 ; by Remark 57.3.9 it does not matter which one) C = KAC,,,

57.

305

SELF-CONTRAVALENCE

with the conditions 51 . I I .2-4. Since A - C * h = h + Ah, Ap, appears in h - C * h as often as p,. If p # Ap, ce-p

xz

-(exp ( A p ) ( h ) ) e A p ,

and = -(exp

ce-Ap

dh))ep

together do not contribute to K in 57.6.1 if compared with 57.5.2, since h * Ah = I (51.1 1.4). Ifp = Ap, thecontribution compared with that in 57.5.1-2 is -exp p(h) = * I . Since 57.5.1-2 were positive multiples of x, determined by these contributions. Whence:

K

is always real and its sign is

Theorem Let G E Alg Lie Com S S be given in ordered maximally compact dressing with respect to C = hAC,,,

where

h = exp 27rih, p(h) = O,+

h E H,,,

for Ap = p

E

W++,

A W + ’ = W”, A * = 1,

AC,,

=

C,, A .

Let A be the top weight of self-contravalentf,

c*A-

=

c 4,p”

(p,

E

W++),

with integral 4,.Put E, = -1 E” =

1

for p,(h) = O for p,(h)

I

=+

if p,

= Ap,.

Then the E in Theorem 57.3 is computed as E=

JJ E?.

fis virtually real if E = 1 and antireal if E = -1. Remark Actually, one may dropall twin factors and,using the classification of

51.17 for every simple summand L, or L,,*,put E,

= -1

except for E,

=

1.

5 1-62.

306

REALITY IN LIE GROUPS AND ALGEBRAS

57.7 It is unfortunate that by the identification of standard and near standard with central and near central types the proof of 57.6 finally rests on the verification in 54.3. Another approach would be to show that up to a positive factor ePpk* * e-pI x does not depend on the choice of p,, . . , ,p k E W + +as long as x, ePplx , . . . , ePPk* * e-p, x is a descent by complete ladders. This has been

shown by H. de Vries (unpublished). D.-N. Verma in his Yale thesis of 1966 (Theorem 4.1, and Remark 4.3) even proved that under the same condition e-Pk * * e-p, is uniquely determined within the associative envelope. 57.8. Antireality and Quaternion Space Theorem An antireal linear representation of real semisimple C, in complex

R can be interpreted as a representation by quaternion linear mappings of a quaternion linear space of half the dimension of R. Proof In the antireal case the expressions

a+PD (a,/3 complex, D the semilinear mapping of Theorem 57.3 with D2 = -1)

may be considered as left-hand operators on R. They form a skew field in which D a = CrD. A conjugation can be defined in this field, a + P D = 8 - /3D

such that

-

uu = on.

Thus the skew field is that of quaternions (Qio). R can be interpreted as a linear space Q over Qio. Its dimension is half that of R . Thef(a) with a E C, are endomorphisms of Q : f(a)( a

+ /3 D) x = a f(a) x + /3 f(a) Dx = a f(a) x + /3Df(a)x = (a+ /3D)f(a)x.

57.9. Historical Note E. Cartan classified the real representations of semisimple Lie algebras [J. Math. Pures Appl. 10, (6) 149-186 (1914) = Oeuvres I 1, 493-5301, mainly by verifications. The explicit formula for E, and the relation between E = -1 and quaternionic representations, was found by H. Freudenthal. 58. COMPUTING

E

FOR SIMPLE LIE ALGEBRAS

A basis of self-contravalent dominant integral elements of H* was given in 56.3. All possible self-contravalent top weights can be combined from them

58.

307

COMPUTINGE FOR SIMPLE LIE ALGEBRAS

with nonnegative integral coefficients. In such combinations E behaves multiplicatively. Therefore it suffices to compute E for the members of that list. The list of fundamental weights of Table F is used also. A,,,: r k coefficients

+

r[+]-k

(k G * I ) expressed in the primitive rootforms has the 1,2,..., k ,..., k ,..., k ,..., 2 , l .

C*(Tk

+

thus in

ri+]-k)=-rk-r[+l-k; rk

+ rl+l-k

- c*(rk

+ rl+l-k)

all coefficients are even: E=

+l.

For I odd, I = 2m - 1, rm has the coefficients 1 2

j’2’ C*n,

m

*

’

., -

¶

* * *9

2 Y

1. j’

= -r,,,; thus E=

A,,,,*:

efficent s

For 1 odd,

2m - 1:

(-1)””. c * r k = -rl+l-k.

1,2,..., k ,..., k ,..., k,

-c

* r k

has the

CO-

..., 2 , l .

Therefore

A unified formula is possible if the signature u ( = 21 - 4j+ 5 for j > 0, = 21 + 1 for j = 0) of the quadratic form is used by which real r2was presented

51-62.

308

REALITY IN LIE GROUPS AND ALGEBRAS

(see 54.1 and 59.7; do not confuse it with that of the Killing form). An easy computation shows fork=!:

E=+I

if o = f l mod8,

€=-I

if a = * 3 m o d 8 .

Dt,,:For k # 1,2,

c*r k = v “k

-c

* T k

k ,

= (k

- 2)(pl

f

p2)

mod 2.

Thus f o r j # 1,2: ~ = + 1 , f o r j = 1:

E

= (-1)”.

For the other basic self-contravalent top weights one must distinguish between “1 odd” and “1 even.” For I even : C*?7,=-77k (k=1,2), 77, - C*Tl = 31p, 772

- C*772 = + ( I -

+ t(l- 2)p, + p3 + 2p4 + 2 ) p , + 34% + p3 + 2p4 +

* *

* * *

+ (1-

+ ( I - 2)p,,

Thus f o r j f 1,2: c=(-l@+’ forj=l:

For 1 odd: only C*(7r,

is left, for which (rl TJ - C*(T,

+

+

if k = 1,2;

~ = + 1

if k = l ,

E=-1

if k = 2 .

772) = -771

- 77,

+ n2)= 0 mod 2.

Thus ~=+1

if k = I , 2 .

2)p,,

58.

309

COMPUTING E FOR SIMPLE LIE ALGEBRAS

In the case of even 1, using the signature (T = 21 - 4 j + 8 as in B,,j, one gets: if k = 1,2 a n d j # 1,2, ~ = + 1 if u = O m o d 8 , if u = 4 m o d 8 .

E=-1

E = +1, if k = 1, E = -I, if k = 2, for j = 1 and 1 even is unexpected.f 7rI and 7r2 are reduced to each other by an outer automorphism of the complex Lie algebra. Clearly, this symmetry is destroyed by urgingj = 1.

Remark The result

For k # 1,2

D,.j,* ( j # 42):

c*7rk = -nk, T k - c*T k = (k - 2)(pl + p2) mod 2. Thus for all cases & =

+l.

For the other basic element one must distinguish between “1 even” and “1 odd.” For I even

+ 7.5) + - C*(7r1+ p1 + p2 mod 2. C*(7rl

7 1

= -7r2

-TI,

7r2) =

7r2

Thus, for all cases, E=+1.

For I odd

c*

3Tl =

-xz,

~ ~ - C * ~ ~ = ~ ( I - l ) ( p l + p 2 ) + p 3 + 2 p 4 + +(l-2)pl.

Thus for k

=

1 (k = 2 is the same case), & =

(-1)(9+j.

Using again the signature u = 21- 4 j + 6 ( j > 2), one can write this as f o r k = 1:

E6,j ( j = 0,1,2): (x4

+

7rJ

- C*(7r4

+

& = + I if

u=Omod8,

& = - I if

o=4mod8.

All coefficients in E=+I.

t

It differs from Cartan’s.

(7rl

+

7r3)

- C*(v,

are even. Thus

7rTTS), x 6 - c*?76

+ r3),

7r2

- C*7r2,

310

51-62. REALITY IN LIE GROUPS AND ALGEBRAS

E6,,,* ( j = 0,2): In every n k - C*Tk the coefficient of p2 and the sum of all coefficients is even: E = +l. E,,, ( j = 0,1,2,3):

For T , , r4,7rS,n7: for n2,rr3,7r6:

E = +I

E = +1

if j = 2,3,

E=-1

if j = O , l .

E, :

E = +l.

F, :

E = fl.

G, :

E = +l.

59. INVARIANT BILINEAR AND SESQUILINEAR FORMS 59.1. Unitary Representations In 35.1 it was shown that every finitedimensional linear representationf of a compact group is essentially unitary. For the infinitesimal algebra Gunof a compact semisimple group this means the existence of a definite hermitean inner product (. .., . . .) in R such that

( f ( a ) x , y >+ (x, f(a)r>= 0

for a 6 Gun.

Expressed in G this reads 59.1.1

(f(4X,Y>

+ (x,f(Cuna)Y)

=0

for a G.

With li in the trunk instead of a and vectors xA,x,, belonging to the weights A,p, this becomes

A ( W h and because of C:,, 59.1.2

XJ

+ =

0,

= -1,

(xA,x,,)

=O

for X f p .

Putting a = ear, one gets from 59.1.1 (f(ea)X , Y> = (x, f(e-a)Y>.

Consider products of branches and put u* = era if u = e,, and (uv)* Then ( f ( 4 X , Y > = (x, f ( U * ) V > . In particular,

= v*u*.

59.

INVARIANT BILINEAR AND SESQUILINEAR FORMS

31 1

(f(4X A , f W xn) = (f@* u) XA, XA>.

59.1.3

Thus, if the weight h has multiplicity 1, thenf(u*u)xA is a nonnegative multiple of XA. The existence of invariant (. . ., . . .) was proved by integration, but an algebraic construction is still lacking. For irreduciblef one should proceed by postulating 59.1.2, taking x A for the top weight h arbitrarily, and x,, of lower weights p, which can be put into the formf(u)xA, by postulating 59.1.3. A crucial point is to prove that f(u*u)xA is a nonnegative multiple of xA. It is rather easy to show that the coefficient is real. The difficulty to ascertain its positiveness algebraically seems still greater than the analogous problem in 57.5.1-2. 59.2-8 G ~ A l Lie g ComSS in C-third dressing and f an irreducible representation of G in R . Iff is not areal with respect to C,C,,,, then semilinear D,Dun,and E, €,, = f l are determined according to Theorem 57.3 by

Df(cU)D-’=f(a), o u n f ( C u n a) Did

=f(a),

D 2= E, Din = €un*

Invariant (i.e., infinitesimally invariant) bilinear and sesquilinear forms will be investigated. Note that they are nondegenerate as soon as they are nontrivial (nonzero) since their radicals are invariant. (. .., .. . .) always means a f(Gu,,)-invariant positive definite inner product on R . 59.2. Uniquenessof Invariants Invariant bilinearorsesquilineart forms, if they exist, are unique up to a scalar factor. Indeed, if, for instance, (. .., . . .) is bilinear and nondegenerate, then some other (.. ., ...)’ can be expressed in terms of (. . ., ...) by (x, y)’ = (Kx, y ) with some linear mapping K; if they are both invariant, for all u E G, (f(u)x, y ) (x, f(u) y ) = 0

+ (Kf(u)x, y ) + (Kx, f(u) y)

=0

for all a E G,

then, by substituting K x instead of x in the first line, one gets

(Kf(u)x, y) = (f(u)K x , ~ )

for all u E G,

which shows that

Kf(u)=f(u) K

for all a E G ;

hence, because of the irreducibility off, K is a scalar multiplier.

t

Linear in the first and semilinear in the second variable.

312

51-62.

REALITY I N LIE GROUPS AND ALGEBRAS

Any invariant bilinear form can be split into an invariant symmetric and an invariant skew part: (X9.Y) =

3((x-,Y) + ( Y , X ) )

+f((X,Y)

-(Y,X)).

Therefore it is either symmetric or skew. For sesquilinear forms analogous remarks are valid. Skew sesquilinear forms become symmetric by multiplication with i. An immediate consequence is the following: Proposition If (. . ,, .. .) is the f(G,,)-invariant product andfis not areal with respect to C, then

(positive definite) inner

(DY, D x ) = < x , y > ; hence, ( x , DY> = E ( Y , D x ) . Proof The new inner product (. . ., .. .)‘ defined by

(X,Y)’ = ( D Y , D x ) is stillf(G,,,,) invariant, since

( f ( a ) x , ~ >+’(x, f ( c u n a ) Y > ’ = ( D Y , Df(a)x>+ . 59.3. G-Invariant Forms Theorem G possesses a nondegenerate bilinear symmetric (skew) invariant

under f iff f is virtually real (antireal) with respect to C,,,,.The invariant can be taken as y rx,y- ( x , D,,,y) where (. .., .. .) is the f(G,,,,)-invariant inner product. Proof The invariance formula

(.mx, Y ) + (x,f(4r) = 0

59.

INVARIANT BILINEAR A N D SESQUILINEAR FORMS

313

applied to weight vectors xA,x,,and h E H yields

(f(h)X h x,,) + (XA, f(h) x,,)

(W + Ah)) ( X A , x,,)

= 0, = 0;

thus,

0

( x A ,x,,)

=

( x A ,x,,)

=0

unless h = -p,

in other words, unless p = C,*,A.

In order for (. . ., . . .) to be nondegenerate, C,*,h must be a weight off if h is so, and have the same multiplicity. Therefore C:n maps weights into weights while preserving multiplicities. Thus Y,f(C,,a) andfare equivalent andfmust be C,,-self-contravalent ; that is, it cannot be C,,-areal. Conversely, letfbe not areal with respect to C,,. Then (x, Y )

(x,

1

Dun

Y>

defines a bilinear form on R which isf(G)-invariant, since (f(a>X, Y ) + (x, f(a>Y ) = (f(a)X,

Dun

Y> + (x, D u n f ( a ) ~ >

= M a ) X, D u n Y>

+ (x, f ( C u n I 0 D u n Y> = 0.

It is evident that (. . ., ...) is nontrivial; it may be symmetric or skew. Now, according to 59.2, (x,

D u n Y ) = Eun(Y, D u n X),

which proves the last assertion, and the one about symmetry.

59.4. G,-Invariant Forms Theorem Gc possesses a nondegenerate sesquilinear form invariant underf

iff its top weight X is equivalent to CC,,h. For simple Gthis means thatfhas to be areal with respect to both C and C,, or with respect to neither C nor Gun. Again using thef(G,,)-invariant inner product (. . ., ...), one can write any nondegenerate sesquilinear form as

Proof

Y ‘x,yl(KX, Y> with nondegenerate K . Its G,-invariance means

X , Y >

+ (KX, f ( C 4 Y ) = 0.

314

59.4.1

5 1-62.

REALITY I N L I E GROUPS AND ALGEBRAS

Kf(CCun a) K-' =f(a),

hence thatfand Y.f(CC,,a) are equivalent and by 44.1 that h and CC,,h are equivalent. All arguments can be turned the other way round. For simple G the equivalence of the other conditions follows by inspection (56.3). Note that a bilinear form invariant underf(Gc) is also invariant underf(G); therefore Theorem 59.3 applies then. 59.5. GJnvariants in Real and Quaternion Space If f is areal with respect to neither C nor C,,, then more information is available via the semilinear D, Dun. Theorem If (. .., . . .) is thef(G,,)-invariant 59.5.1

inner product, then

Y r x , y i ( X 1 Dun DY)

is an f(Gc)-invariant sesquilinear form, which can be made hermitean by a suitable norming of D ; it is hermitean iff (D,,D)2 = 1. If E = 1, its restriction to R , coincides with that of the bilinear form of 59.3; it is real and symmetric if E,, = 1 ; it is imaginary and skew if E,, = -1. If E = -1, then considering R a s the quaternion linear space Q (see 57.8), one gets an f(Gc)-invariant quaternion sesquilinear form (. . ., .. .), 59.5.2

(x,Y) = P<X. DD,,y>

+ ,f?D(D,,x,y>

with p = -~,,f? (e.g., = 1 for E,, = -1, ,f? = i for hermitean, if E,, = 1, it is skew hermitean.

E,,

=

1). If

Proof Thef(G,)-invariance of 59.5.1 follows from

f(CC,,a)

Furthermore, because of (CCJ2

=

D-' D;Af(a) Dun D.

=

1 and the irreducibility off,

(DunD ) 2 = scalar y ,

E,,

= -I,

it is

59.

INVARIANT BILINEAR AND SWQUILINEAR FORMS

315

hence (DDun)2 G ( D u n D)’Dun=y, and by multiplying both equations

w =1. By Remark 57.3.9 one may use /3D instead of D, with /?’ = y-’. This allows one to suppose (DunD ) 2= 1. Thus, because of 59.2, (X,DunDY) =~un = ~ u n ~ ( D D u n x ,= by

Y r x . y 1 ( K x , Y>, which should be calculated for weight vectors x , y. According to 59.4.1, if x, belongs to the weight p, then Kx, does to the weight A p . By 59.1.2 ( K x , , ~ , ) vanishes unless A p = v. Therefore one can restrict the computation to (Kx,, x;,). (The prime on x;, accounts for the circumstance that by chance Ap may equal p, though there might be several independent p-weight vectors.) Starting from the top weight A, one has p =A -

c

P V P V

and a linear combination u of products of e-p (p E W++) such that x,

=f(u)XA.

Applying 59.4.1 and taking CC,, e-pv= AAe-,,

= -E,

e-Apv

into account (see 57.6), one gets Kf(e-Ap~)K-’

-&v

=f(e-pv);

therefore Kf(Au) K-’ =f(u).

(-E”)’”

If AX = A, then, after having replaced K by a suitable multiple (real if the sesquilinear form was already hermitean), KxA = xA, x,

=f@) XA = =

for a suitable choice of x,,, Therefore Kx,

=

n(-&,)” Kf(Au) n(-~,)”’ Kx,,

XA

which, however, coincides with x, if A

n (-E,,)~~X,

=

for A = 1.

This then completely determines the nondegenerate hermitean form

Y r,.,*lWX, u>.

1.

59.

INVARIANT BILINEAR AND SESQUILINEAR FORMS

317

lffis areal with respect to neither C nor C,, and D , Dunare normalized as in Remark 59.5.3, then in the case Ah = h K

=fDD,,,

and, if necessary by replacing D by -D, even K

= DD,,.

59.7. Examples For the lowest dimensional linear representations of G E Alg Lie Corn S S S one gets the following results by which the assertions of 54.1 come true.

r l ( A l , , ) : In the notation of Section 16, w I , ..., w l + ,are the weights, with Kx, = x , for u G j , Kx, = -x, for orthonormal weight vectors x I , . . ., u >j . The invariant hermitean form has the signature 2 j - (I + 1). By comparing the dimensions, one verifies that the group consists ofalllinear mappings with unit determinants that leave this form invariant. By 59.3 no invariant form exists. Comparison r I ( A 2 m , O , * ) , rl(AZm-l.m.*): of dimensions shows that the group consists of all real linear mappings of R, with unit determinants. n,(A2m-l,o,*): Since E = -1, the representation takes place in quaternion m-space. All its automorphisms form a 4rn2-dimensional group; those with unit determinants produce the correct dimension. (The determinant is defined as a real number after waiving the quaternion structure.)

r2(Bl,,): E,, = E = 1. Weights *w,,, 0 with vectors x+,,,xO.t Because of Proposition 59.2, one may suppose that Dunxu= x-,, since Dunpermutes the eigenvectors as C:, permutes the weights. Then, in the normalization of 59.6, D = KD,,. F o r j > l : Kx+,=xtu if l , t u < j ,

Kx+,=-xt,

if j G u < l ,

KxO = -xO.

+

is spanned by ix,, T X , ?x-, (1 < u <j), TX, - Tx-, ( j G u G I), which make ( K x , ~ 0, < 0. The invariant quadratic form has the signature -1 (2j - 2) - 2(1- j + 1) = 4 j - 21 - 5. RD

Forj=l:

+

Kx+,=xt,

if

~GUGI,

KxO = -xO.

RD is spanned by ixo,T X , + f x - , (1

G u G I).

Signature 21 - 1.

t Here, and in the remainderof this section, the signs in the subscriptsare used dependently.

318

5 1-62. REALITY IN LIE GROUPS A N D ALGEBRAS

Comparing dimensions, one verifies that the groups are the 1-components of the full invariance groups of these forms. rl(Cl,,),j < I: E,, = E = -1. Weights fw,, with vectors x+,. One may suppose that Dunxu= x-,,, Dunx-,= -xu.

Kx,,=x,,

for 1 Q v ~ j ,

Kx,, = -x,, for j < v Q 1. The x , (v > 0) form an orthogonal basis over the quaternions. The invariant quaternion hermitean form has the signaturej - (I -j). Comparing dimensions, one verifies that it is the full invariance group. T ~ ( C ~ , E~, )=: -1, E = 1. By 59.3 an invariant real skew form on R,,. The full real symplectic group. r 3 ( D i , , ) , j #1,2: E, = E = 1. Weights fw, with vectors xi,,. One may suppose that Dunx,= x-,,.

Kx+, = x*,, Kx,,

= --xi,,

for 1 Q v Q j - 2, for j - 1 Q v Q 1.

Computations as in n2(B1,,) lead to a quadratic form with signature 2(j- 2) - 2(1- j + 2) = -21

+ 4j-

8.

~3(Dl,l): E, = 1, E = -1. An invariant skew hermitean quaternion form. Comparing dimensions shows that the form characterizes the group. With x,, and D,, as before, the form as indicated in Theorem 59.5 becomes ( x , , ~ ,= ) i for v < 1, ( x , , x , ) = -i. As a matter of fact, all nondegenerate skewhermitean forms on quaternion n-space are equivalent under automorphisms of quaternion n-space. This is an elementary fact,t but it can also be derived from the equivalence of the linear Lie groups leaving such forms invariant.

7r3(Dl,,,*):

1. A basis as in Dl,, shows ifj23: Kx*y=xiv for l ~ v ~ j - 2 ,

E,

=E =

Kxkv = -x+,

for j - l < v < f ,

K X , ~= - x T i ; ifj=O:

Kx,,,=x+,

for 1 g v ~ l 1, -

K x , ~= x T I .

+

(1 Q v Q j - 2), rx,, - ~ x - , (, j - 1 Q v < I), RD is spanned by r x , ?xu i(x,+x-,)forj>3; by~x,+?x-,(l ~ v < I ) , x , + x - , , x l - x x _ , f o r j = O .

t J. Dieudonnk, Trans. Amer. Math. Soc. 72,383 (1952).

59.

INVARIANT BILINEAR AND SESQUILINEAR FORMS

319

The signature of the quadratic form is forj>3: for j

= 0:

(2(j-2)+ 1)-(2(1-j+2)-

1)=4j-21-6;

21 - 2.

Again, the quadratic form characterizes the group. 7](E6,1): dim R = 27. Areal both with respect to C,, and C. There are, respectively, 1,16,10 weights of the form 7rI - 0 p, - * *, 7rI - 1 * p1- * 7r1 - 2p1 - * * * . Signature of the invariant hermitean form 11 - 16 = -5.

-

-

a ,

Tl(E6.2): dim R = 27. Areal both with respect to C,, and C. There are, respectively, 6,15,6 weights of the form 7r1 - 0 p2 - -,7rI - 1 * p2 - ., 7rI - 2 * p2 - * * . Signature of the invariant hermitean form 12 - 15 = -3.

--

7r,(E6,0,*),T1(E6,2,*):

--

By 59.3 no invariant bilinear form.

~ 2 ( E 7 , 0 ) , ~ 2 ( E 7 , 1 ) : dim R = 56. E, = E = -1. There are quaternion hermitean forms. The opposite of a weight is a weight; all are simple. As a quaternion orthogonal basis one may choose vectors of weights in which p2 has a positive coefficient. For

7r2(E7,J the signature equals the quaternion dimension 28. For

there are 11,16,1 weights of the form 7 r 2 - 0 . p 1 * -,7r2 - 2p1 - * * in which pI has a positive coefficient. The signature of the form is -4. ~2(E7,1)

7rz

-

. . a ,

-

- 1 pI,

m2(E7,2), ~ z ( E 7 , 3 ) :

dim R = 56. ,E,

= -1, E = 1.

Skew real bilinear forms.

7rl(E8,,): E,, = E = 1. The invariant quadratic form is the Killing form, since the representation is adjoint; for signatures, see 52.6.

T ~ ~ ( F ~l(F4.2): ~ , ~ ) , dim R = 26. E, = E = 1. The invariant quadratic form has the signature 1 - 8 8 - 8 + 1 = -6,5 - 14 + 7 = -2. (The weight 0 has the multiplicity 2.)

+

T ~ ( G ~ ,dim ~ ) R: = 7. E, signature 2 - 3 2 = 1.

+

=E =

1. The invariant quadratic form has the

59.8. Real Invariants in the Areal Case Iff is not virtually real with respect to C,one may still ask for invariants under mf(GC). However, this opens no new views. wfis obtained fromfby waiving the complex structure of R. This procedure transforms a bilinear or sesquilinearf(Gc)-invariant form into a bilinear one, invariant under rof(Gc); by separating the real and imaginary parts, one obtains real invariant forms.

320

51-62.

REALITY IN LIE GROUPS AND ALGEBRAS

Conversely, if (. .., ,..) is a real bilinear invariant of mf(G,), then y rx,,.i(ix, y), y rx,yi(x,iy),y rx,,.i(ix,iy) are also invariant, and so is

Y rx.,.-4x,

Y ) k Ox, iy) - i(ix, Y ) k i(x, iv)).

It is bilinear for the lower signs and sesquilinear for the upper. It is impossible for both to vanish unless (. .., ...) does. Theorem All bilinear invariants of mf(GC)in the not virtually real case can be obtained from bilinear and sesquilinear invariants off(G,) by waiving. In the antireal case, however, they exist only iffis not areal with respect to C,,, and then all real bilinear invariants are assembled in the quaternion sesquilinear form 59.5.2. 59.9. Historical Note Though many particular results were obtained by

E. Cartan, the general methods and theorems of this section are new.

60. MINIMALLY COMPACT DRESSING G E Alg Lie Com SS in ordered C-third dressing on H; thus CH = H = C,, H. For notational convenience G is assumed to be the infinitesimal algebra

of a linear Lie group G. Sometimes C-third dressing occurs on a trunk H'; then C:, plays the role of C,,, and he is replaced by he'. 60.1-8. he- Notions 60.1. Definition A maximal abelian subset of GCWhe is called a he-trunk. The notion depends on the choice of C,,. Since the elements of GC,he= iG,,,, are ad-pure, any he-trunk is ad-pure,

hence contained in a trunk. Proposition Let A be a he-trunk.

(1) The centralizer in G, of A is contained in A + G,,,,,. (2) Any maximal abelian subset D between A and Gc is a trunk of G,. (3) Any abelian subspace between A and Gc has the form A + B with

B c Gc,un(4) If there is a he-trunk in H, it is (5) Every C-invariant trunk containing A is C,,-invariant. (6) Two C-invariant trunks containing A are conjugate by means of some u E Gc,,, centralizing A.

60.

MINIMALLYCOMPACT DRESSING

32 1

Proof

+

(1) Splitting a E Gc in a = s u with s E GC,he,u E G,,,, (see 51.5.8-9), one gets GC,un,

and if [a,A]= (0)

GC.he,

+

[s u , A ] = (0);

hence, [s,A ] = [u, A ] = (0). The maximality of A causes s E A, which proves (1). (2) Because of (l), D = A + B with some B c Gc,,,; A and Bare ad-pure, and so are D and its complex extension DComin G; D,, is C-invariant. If x centralizes DCom,then Cx, +(x Cx), and (1/2i)(x - Cx) do likewise. The latter are C-invariant, and because of the maximality of D they belong to D. Thus x E DCom,which proves ad-pure Dcomto be a trunk of C and D to be a trunk of Gc. (3) Follows from (1). (4) Follows from (l), if A is assumed in H. (5) Follows from (3). (6) According to (l), the centralizer of A in Gc takes the form A + 2 with Z c GCmun. Z generates a compact group, namely, the 1-component of the centralizer of A in Gc,,,. Therefore by 35.7, 2 = 2,+ Z , directl-y,where 2,is central and 2,is semisimple. 2,generates compact 2,. Let Hi (i= 1,2) be two C-invariant trunks of G around A. Since they centralize A, their C-restrictions Hi,c are in A + 2;hence they are centralized by 2, which is in the center of A 2.Being trunks they must contain 2,. Therefore

+

+

Hi,, = A

+ 2,+ Hi,c n 2,.

The (Hi n Z1),,, are still ad-pure with respect to semisimple Z,,Com. Clearly, they are maximal in this respect and consequently are trunks of 2, as are Since ZIgenerates a compact semisimple group Z,, accordthe Hi n 2,of 2,. ing to the conjugacy theorem they are conjugate by means of some u €2,. So are the HiPC and finally the Hi. 60.2. Definition An abelian subset A of Gc is called i-compact if exp iA is relatively compact. Clearly, a he-trunk is i-compact, and because of 60.1 it is even maximally so. The converse is also valid :

322

51-62.

REALITY IN LIE GROUPS AND ALGEBRAS

Proposition A is maximal i-compact if and only if A is a he'-trunk for a suitable unitary semimorphism C:,-in fact, for any C-third dressing semimorphism on any C-invariant trunk containing A .

By 35.9 compact exp iA, hence iA, is contained in a trunk. Thus A is centralized by some regular x E G and also by Cx and T X + TCx, which is C-invariant and, for suitable T,regular. A is contained in its trunk, which again is C-invariant. Let H' be any C-invariant trunk around A . With respect to this H' a C-thirddressing C;, is chosen. It serves to define Ghe,. Since A c Ghe,,there is a he'trunk B 3 A . Now B is i-compact, and A is maximally so. Thus A = Band A is itself a he'-trunk.

Proof

60.6. Definition a E GC,heis called he-regular if rank f for x E GC,he takes its maximum at a, equivalently, if rank 3 is constant in some neighborhood of a within GC,he. Proposition Every element of GC,heis in some he-trunk. A he-regular element of GC,heis in one he-trunk only. Proof The first assertion is obvious. Let h be he-regular and A its centralizer

within GC,he. Then %A= (0) and 6 A c A for a E A . Now he-regularity of h prevents 61Afor a in A near h from having nonvanishing eigenvalues. On the other hand, because of A c GC,he,any a E A is ad-pure. Hence 6 A = (0) for a E A . This shows that A is a he-trunk, and the only one containing h. Remark If A is a he-trunk and rank (x E A ) takes its maximum at a, then the centralizer B of a in GC,he coincides with A . Indeed, A c B, and by the previous argument, B centralizes a neighborhood of a in A , whence A itself, which however was chosen maximal abelian in GC.he*

In 60.12 even the regularity of such an a will be proved. 60.7. Definition u E W* is called a he-nil-rootform if a(Hc,J

= (0).

Proposition a E W* is a he-nil-rootform iff a = CC,,a. Proof If a is a he-nil-rootform, then for all h E Hhe ((1 - CC,,)

.)(a)

R) = a(( 1 + C )1)= 0,

= a((1 - CC,,)

and conversely. Since (1 - CC,,) a is linear, vanishing on Hhe and on H imply each other.

60.

323

MINIMALLY COMPACT DRESSING

60.8. Proposition If h E H C , h e is he-regular, then a(h) # 0 for every a E W* that is not a he-nil-rootform. Indeed, if h is regular, then rank 1 is also constant in a neighborhood of h within HC,he. The converse is true if HC.he is a he-trunk. It will be verified in the course of is called he-semiregular if a(h) # 0 the proof of 60.10. Meanwhile, h E for all non-he-nil-rootforms a. 60.9. Criterion o n Trunks Containing a he-Trunk If a is a he-nilrootform, then CCune,= *e,, since (CC,,J2= 1 .

Proposition H (= CH = C,,H) contains a he-trunk iff CCunea= e, for every he-nil-rootform a. Proof H containing a he-trunk means that

thus that U E G C , h e and [ U , H c , h , ] mod H C , h e the sum of expressions GC,he,

=

(0) implies U

is maximal abelian in Any U E C C , h e iS

E HC,he.

(1 - CCun)(Taea + ~ae-or)

60.9.1

with T , # 0. For h E HC,he,[h,a]becomes a sum of expressions 60.9.2

a(h)(1

+ CC,,)

(7,

e,

- 5,

e-,).

Thus H C , h e is a he-trunk iff the vanishing of 60.9.2 implies that of 60.9.1. This implication can be broken into the conjunction of two implications: a

a he-nil-rootform

and [ a non-he-nil-rootform

A

(1

--f

(1 - CC,,)(T, e,

+ ~ C , , , ) ( Te,, - C, e-,)

+ T,e-,)

=

= 01 --f ( I - CCun)(7,e,

0,

+ Q, e-,)

= 0.

Now the second implication is valid anyhow, since if the antecedens were true, then by linear combination (1 + CCun)TOea = 0, hence CC,,a = a which by 60.7 contradicts the first part of the antecedens. Therefore H being a he-trunk is equivalent to.the validity of the first implication for all T , # 0, which is just the condition to be proved. of Trunks Containing a he-Trunk Clearly the property that H contains a he-trunk is G,-,,-invariant. The converse is also true : 60.10. G,,,,-Conjugacy

Proposition If H contains a he-trunk, then every C-invariant trunk containing a he-trunk is a G,,,,-conjugate of If.

324

51-62.

REALITY IN LIE GROUPS AND ALGEBRAS

Proof By 60.1.4 HC,he can be assumed to be a he-trunk. By 60.1.6 the Cinvariant trunks around HC,he are Gc,,,-conjugate. Therefore it suffices to

prove that #Hc,heruns through all he-trunks if u runs through Gc,,,. If for a non-he-nil-rootform CL one applies d with a = exp( I

+ CC,,)

(7,

e, - TKe-,I

E Gc,un

to b E HCahe, then one obtains d b = h - a(b)(l -CC,,)(~,e,+?,e-,)+

Suppose that bo E H C , h e is he-semiregular. Then a(ho)# 0 and according to 60.9.1 the - ccUll)(7K

eK

+ a'

span G C , h e mod & h e . This means that the &/I cover an b,-neighborhood in C C , h e if b runs through an b,-neighborhood in & h e and a runs through a l-neighborhood in Gc.,,. Hence rank f (which is a Gc,,,-invariant) is constant in a neighborhood of Lo within G C , h e , which shows that Lo is he-regular and confirms the announced converse of Proposition 60.8. y=

U a ~ G ~ . UdHC.he a

is closed in (&,he because of the compactness of Gc,,,. Every he-regular point of &,he, and therefore any Gc,,,-conjugate of such a point, was proved to be an interior point of Y with respect to GC,he.Therefore the boundary points of Y are images of such b E in which at least one non-he-nil-rootform vanishes. An argument as in 31.9 shows that the boundary of Yhas codimension > 3. So it does not decompose GC,he, and Y coincides with GC,he. This proves that every element of G C , h e is a G,,,,-conjugate of an element of H C , h e . Now let A be an arbitrary he-trunk and 6 E A such that rank .f for x E A will reach its maximum a t 6. For some u E G,,,, it happens that #-I6 E &,he. Now #-'A is again a he-trunk and, according to Remark 60.6, the Hence &.he c t i - ' A , thus H C , h e = #-'A, thus centralizer of #-'b in A = nffc,he. 60.13. Gc-Conjugacy of Trunks Containing a Maximal i-Compact Subset Clearly, the property of Hcontaining a maximal i-compact subset is &--invariant. The converse is also true: Proposition ccacts transitively on the set of C-invariant trunks of Cc containing a maximal i-compact subset. Proof Let H, H' be such trunks. C,, is chosen as a C-third dressing unitary semimorphism with respect to H. Then by 60.2 Hcontains a he-trunk. By 51.7 there is a Gc-conjugate H" of H' such that ZH= H" for some u E Gun.H"is

still C-, and C,,-invariant and contains a maximal i-compact subset. By Proposition 33.13 C,, counts as a C-third dressing unitary semimorphism for H"; hence by 60.2 H"also contains a he-trunk. Therefore 60.10 can be applied

60.

325

MINIMALLY COMPACT DRESSING

to the effect that H,H" are G,,,,-conjugate. This proves that H,H' are G,conjugate. 60.14. Real Rank Definition The dimension of a maximal i-compact subset of C , is called the real rank I,,, or I ' , of C,. By 60.1 and 60.13 it does not depend on the choice of the set. Evidently : Proposition The real rank of Gun is 0, that of G,, equals the rank 1 of C .

60.15. A Procedure For a trunk H = CH = C,, H (CC,, = C,, C ) that does not contain a maximal i-compact subset of C, the condition of Proposition 60.9 is not fulfilled. Therefore for some he-nil-rootform a

CC,, e,

= -ea,

hence, CC,, e-,

because of

= -e-,,

CC,, ha = ha.

On the evidence of such an a there is a Procedure of changing H into a trunk H' = CH' = C,,H' such that H n H' is the kernel of u (which contains HC,he) and ih, E H,,,, is replaced by some h' E G C , h e . By this step the dimension of the i-compact part HC,heof H , is raised and that of the torus part is lowered by one.

U,

= exp

+

~ ( 0 , 0-,),

T

= 377(2(cr, a))-'I2i,

leaves the kernel of a invariant. According to 33.1,

u,2

=

s,,

where S,

u

S, ha = -ha,

= -a,

S, e,

= e-,,

Moreover, Cun u a Gin'

since Cunea= -e-, and

T

is imaginary;

cu, c-' = u,' since Ce,

= e-,;

= ua,

=

C u , ha = U, S,' Ch,

since Ch,

=

u, s;',

thus =

U, h,,

h-,. On the other hand,

C,, U, ha = -U, ha.

S, e-,

= e,.

326

51-62.

REALITY I N LIE GROUPS AND ALGEBRAS

Therefore uu hm

GC.he

Moreover, since U , leaves the kernel of a invariant, H ’ = U , H is a new trunk with the required properties. 60.16-17.

Minimally Compact Trunks

60.16 The time has come to recall the definition of minimally compact trunk (51.8). The procedure in 60.15 shows that if a C-invariant trunk does not contain a maximal i-compact subset of C,, its torus part can be diminished so it was not minimally compact. If it does contain such a subset, the dimension of its torus is 1 - l,,, which is a lower bound for any torus in G,, thus the trunk is minimally compact. 60.17.

Some of the preceding results are summarized.

Proposition The trunks containing a maximal i-compact trunk are just the minimally compact trunks. Conjugacy Theorem of Minimally Compact Trunks The C-mini-

mally compact trunks are G,-conjugate. Definition A C-thjrd dressing on minimally compact H = C H is called a minimally compact dressing.

60.18. Real Ordered Dressing A special ordered dressing is adapted to the study of nonmaximally compact trunks. Definition An ordered dressing is called real (with respect to C) if

60.18.1

aE

W + AC * a < O + C * a = - a .

Rememberthat C*actsas+l on HZ,,,andas-I oniHZ,,,.Itisnotrequired that C and C,, commute, though their restrictions to the trunk H = C H = C,, H necessarily do. The existence of a real order is seen as follows. Take ho E H,, such that a(hJ # 0 for all a E W*, and for every a E W*\iH&,, Iadho)l > Ia*(ho)l

if u = u I+ u2, a I E HZ,he,a2 E iH:,,,,; now take the minimal partial order contained in the partial order as constructed in Proposition 25.2.2.

60.

327

MINIMALLY COMPACT DRESSING

Proposition In a real order CC,, acts on (suitably numbered) Wt+ as follows :

CCun p j

for j = 1,.. ., k,

= pj

mod

CCunpj=-pj

CC,,p,

= -pf+n,+l-j

* *

for j = k + l ,

pk

mod p l ,

. . ., p k

for j = m

..., m,

+ 1, .. ., I ;

or, eliminating C, c * p . = -p

for j = 1, . . ., k,

.

c*p j = p j mod

pl,

. . ., p k

for j = k

C*pj=pft,tl-jmodp,, ..., pk

+ 1, . . ., m,

for j = m + 1,

..., 1.

CC, behaves as -1 on HC*,heand as +1 on iH:,,,,. On the given order let p l , . .., Pk be the primitive rootforms invariant under CC,, and p k + l r ..., pi the others; CC,,p, k . Let N be the linear mapping CC, reduced mod P I , .. ., p k , Then N 2= 1 and all matrix coefficients of N on the basis p k + l , . . ., pi are nonpositive integers. This can only happen if in every row and column of -N there is not more than one nonvanishing coefficient, which has to be 1. Consequently, -N is a permutation matrix and the matrix of CC,, on the basis W++can be put into the form Proof

0 60.18.2

-; -v 0

with

:.

V=(

1

I).

0

This proves the assertion.

60.19. How Far Is C Determined by Its Behavior on H Theorem If H = CH, then C is determined up to inner isotypy by its behavior on H and on the branches (in third dressing with respect to H ) e, with C * a = -a. Remark For maximally compact trunks the proposition carries no new

information. (See 51.12.) For minimally compact trunks it states that C is completely determined by its behavior on the trunk alone. Proof C, C' are taken to coincide on H and on the branches e, with C* a = -a; C ,, has been taken with respect to H and with CC,, = C,C. If used,

G is supposed to be centerfree. The first step is to replace C' by inner isotypic C" such that C"C,,, = C,,C" is also valid.

328

51-62.

REALITY IN LIE GROUPS A N D ALGEBRAS

Since C, C ’ are the same on H, there is an h E H , h = exp h such that

C’= hc,

60.19.1 N

and because of C’* = 1, hCh = 1 ;hence 60.19.2

a(h

+ Ch) = 0 mod 27ri

for a E W*.

One tries

C”=h0C’h,l

with ho=expho,

To have

h0€H.

C ” c,, = c,, C “

means

hOhCh,~c,, = c,,Kohch,l

or, equivalently,

C,,(h, h(Ch,’))

= ho h(Ch,’),

in other words, imaginary for all a E W*. a(ho - Cho + h) This can be satisfied by putting ho = -+h. Indeed, then a(h0 - Cho

+ h) = $a(h + Ch)

which by 60.19.2 is even = 0 mod Ti. Henceforth it is supposed that C’ c,, = c,, C ’ ,

cc,, = c,, c, thus,

hC,,

= C,,h,

C,, h = h,

h E Hun.

So 60.19.2 can be written as 60.19.2’

a(h

+ CC,, h) = 0 mod 2ri

for

a E W*.

Now the identical behavior of C, C‘ on eOrwith a = CC,,a is used. This means that he, = e, for such a ; thus a(h) = 0 mod

27ri

for a = CC,, a.

On a real order (60.18) this amounts to p,(h)

=0

mod 27ri

for j

=

1,

. . ., k.

Without impairing the relation h = exp h, one may even change h such that 60.19.3

p,(h)=O

for j = 1,

..., k.

60.

329

MINIMALLY COMPACT DRESSING

In H,, a basisp,, ...,pl is taken, correlated with that ofp,, .. ., pI:

On that basis the action of CC,, is described by the transpose of the matrix 60.18.2. Then, with imaginary coefficients cj, 1

m

h

c

=

CjPj+

j=k+l

c CjP,.

m+l

Further, 60.19.4

CC,,pj

= -pj

CC.,pj

=-

for j = k + l ,

P ~ + ~ + , - for ~ j= m

..., m,

+ 1, . . ., 1,

and using 60.19.2' 60.19.5

c j = c ~ + ~ +mod , - ~ 27i

for j = m

+ 1, . . ., 1.

To prove that C' and C are isotypic, a g E Hunsuch that g'CC,,g'-' withg = exp g is required. This means thatg has to fulfill 60.19.6

a(g - CC,,g)

= a(h)

mod 2xi

C

j=k+l

E

hCC,,

W*.

&(I-m)

m

g=+

for a

=

cjPj+

C

j=m+l

CjPj

satisfies this, since (60.19.4) 1(I-mJ

m

-CCung=3

C j=k+l

cjPj+

C

j=m+l

cjPt+m+l-j

substituted into 60.19.6, if 60.19.5 is taken into account, verifies 60.19.6. 60.21, A Procedure The contents of the present section is a procedure for changing an arbitrary trunk H in C-third dressing into a minimally compact trunk. C,, is defined on H , H = CH = C,, H , CC,, = C,, C. Let B be the subset of W* defined by a

EB

CC,, ea = -ea.

t)

Then for u E B, also CC,, a = a, CCune-a= -e-,. Let B' be a maximal subset of B such that

330

5 1-62.

REALITY IN LIE GROUPS AND ALGEBRAS

When using the Ua and Sa from 60.15, one notes that the Ua with a E B' commute with each other because of 60.21.1. One defines

u=nua, s=n a sa. UEB'

Then from 60.15

c,, uc;,'

=

CUC-'

u,

=

u-' = us-'

For the actual computation it is more convenient to change the semimorphism up to isotypy than to change the trunk. Thus

C'= u-' cu = s-' c

60.21.2

is an isotypic semimorphism with still C'H = H a n d C' C,, Since Sa = -a, Sea = e-a for a E B', one gets

C' C,,

= C,,

C'.

C' C,, ea = -e-a.

u = -a,

Under some additional assumptions H will come out as a minimally compact trunk with respect to C' (as will UH with respect to C). To prove this one must consider any YE

w*,

y=C'C,,y

and ascertain that 60.21.3

C'C,,

e,,

= e,,,

which will show the minimal compactness of H, thanks to 60.9. By 60.21.2

thus,

CC,, maps the first member into its opposite while leaving the second invariant; hence 60.21.4

Since the members of B' are mutually orthogonal and consequently linearly independent, it follows from the last equation that (Y,B') = (0).

60.

331

MINIMALLY COMPACT DRESSING

If Wpossesses no ladder of length 2 except those passing through 0, one may continue with the conclusion that 60.21.5

a

E B'

-+ [e,, e?,] = 0,

hence 60.21.6

Se,

= e,.

60.21.4-5, together with the maximality of B', show that

CC,, e,

= e,

from which by 60.21.2 and 60.21.6 C'Cuney= S-' CC,,e,

= S-le,, = e,.

If, however, Wdoes possess nontrivial ladders of length 2, it can happen that

[ear,e*,l f 0

though ( y , a) = 0,

namely, if a is the midpoint of its y-ladder. Nevertheless, the former conclusion holds good as long as 60.21.7 for y B'.

E

W*, (y, B') = {0}, the midpoint of no y-ladder belongs to

60.22-23. Toward Computing Minimally Compact Trunks

60.22 What matters according to 60.19, is to know the behavior of C' as obtained in 60.21 on H.H: splits under C' C,,, into HZ,+and H:,-, which belong to the eigenvalues +1 and -1 of C'C,,,. (Also H:,+ = iH$,,,,, H:,- = H:,,hc.) In the subspace spanned by B', C' C,,,behaves as -CC,,, in its orthoplement as CC,,,. If maximally compact C-dressing was the point of departure, then on H the splitting of Hst under CC,, is into the +I-eigenspace (1 CC,,,)H,, and the -1-eigenspace (1 - CC,,,)H,, (for inner types = 0); B belongs to (1 + CC,,,)H:, as does B', whence:

+

Proposition If H is a maximally compact trunk with respect to C, and isotypic C' has arisen by the method in 60.21 such that H is minimally compact with respect to C', then, under C-third dressing and with the notation of 60.21, the (-1)-eigenspace of C'C,,, on H: is spanned by B' and (1 - CC,,,)H:, which are linearly independent. Its dimension is the real rank of Gc.

60.23 The procedure of 60.21 will actually be performed on all types of G E Alg Lie Rea SSS, starting from the classification dressing of Sections

51-62. REALITY IN LIE GROUPS AND ALGEBRAS

332

52-53. Since, according to 60.19, it is enough to know C'C,, in H:,

a basis of the (-1)-eigenspace of C'C,, is indicated. Twin and compact types may be disregarded. Basis of the (-1)-Eigenspace of CC,, in H* under Minimally Compact Dressing

(See a somewhat different form in Section 75.)

-

AIJ, j < t(l+1): pj, pj-1 +p, +p,+1,. . ., P I + * * +pj + * Or, otherwise: o,- w ~ +w,-~ ~ , - w , + ~ , ..., w 1 - w2,. Real rank :j . B1,j, 1 <j < #l+ 1: p j , p J - l + p j + p j t l , . . .,p2+ * + p j + 2 ~ 1pj, 1 + p, pj+l+ 2pj+2 + * * * pj +2p,t 1 * * .

+

+ +

+

+

P2

* * *

+ P2,-2 +

(

* *

+p2,-1,

. . . +p2j-2,

+ 2 ~ 1 -,

* *9

for 1 # 2 j - 2, for I = 2 j - 2.

2p2,-1 2PI

Or, otherwise: w,-~ & wJ, w j - 2 f o , + ~..,., w I f w 2 j - 2 . Note that no intermediate rootforms show up. Real rank: 2 ( j - 1). Bl,j,j>-31+1: '

+pj+

fI,PI-l+pj+pi+l,...,P21-1+

+ p1 + p t , pj + 2pj+1 +

*

*

* *

'

+2p, +2p1,

*

.., p2j-l+

*

* * *

+pI,pzj-t-l+ +p1+2p1. *

a

Or, otherwise: w j - l w , , ~ , - ~ i w ~..., + ~w2J-I-I , f wl,w2j-l-2.Thereisone intermediate rootform, p2j-l+l + . * * + p I , namely, in the y-ladders with y = k(pk * * * p1 p l ) ( k > 2 j - 1 + 1); however, this y is orthogonal neither topk+ *"+&j-k+2&j-k+l+ * ' * f 2 p ~ f O r k ~ j n O r t o p z j - "l 'f + p l + 2 p l for k >j . Real rank: 21 - 2 j 3.

+

+ +

+

Bl,l : p l or, otherwise, wI.

Real rank: 1 . Cl,j,j,Pm pm+1, Pm-1 Pm+2, . * *, PI * * * P2m. Or, otherwise: w1 w ~ ~ ( i+= ~1, 2, - ~ ., m),w, a.9

+

w1-

+

+

+ Pm + pm+l +

..

-

...,

W[.

Real rank: 1. Pi-p2mn-r (i=1,2, **-,m11, pm, pm-1 +Pm +Pm+l, * PI +... + Pt. Or, otherwise: oi- w2m+l-t( i = 1,2, ..., m - l), w, - a m + [ , -

Azm-l,m,*:

s - 3

...,w1 - w1.

Real rank: 2m - 1. D1,0,*:

P I - P2.

Or, otherwise: Real rank: 1.

wl.

-

+

+

+

* Dt.,.*, 3 G j G 3r + 2: PI, PI-1 + PJ P,+I, * - 7 p3 + P2,-3, P, + +,+I * * ‘ +2pt +PI +p2, * * * ? p3 + * * ’ +pZ,-3 +2p2,-2 + * ’ ’ +2pl +PI p2, PI -pZ. Or, otherwise: w , - ~f w , - ~ , f w,, ., w I f w2j-4, w I . Real rank: 2 j - 3.

..

Eb.O.*:

+

PI - P39 P4 - Ps.

Real rank: 2. E6*2.*:

1 1 00000’ 01210’

1 1 11211 ’ 12221 ’

- p3i

p4 - ps.

Real rank: 6. 60.24. Historical Note Minimally compact trunks for the different real types were found by E. Cartan [Ann. Ecole Norm. 44, 345-467 (1927) = G?%ures I 2, 867-9891. He also stated their conjugacy. The criterion of 60.9 seems to have been formulated first by H. Freudenthal, though it was prepared

61.

REAL SEMISIMPLE LIE GROUPS AS PRODUCTS

335

by investigations to be dealt with in the next section. The theorem in 60.19 was proved by H. de Vries. The procedure of constructing minimally compact trunks (60.21) is possibly the same as that by which E. Cartan obtained his minimally compact trunks.

61. REAL SEMISIMPLE LINEAR LIE GROUPS AS PRODUCTS OF MAXIMAL COMPACT AND SOLVABLE GROUPS 61.l.Theorem G E Gru Lie Lin Com SS; C is an involutory semimorphism, C,, is a unitary one, and CC,, = C,,C. Then there is a closed solvable Lie subgroup X of Gc such that every g E Gc splits uniquely as g = ux with u E Gcnun,x E X . 61.2 The splitting is first performed on the infinitesimal algebras. The trunk H of G is chosen minimally compact, H = CH = C,,H, in real order; this means that if a E W,u> 0, CC,,a # a,then CC,,a < 0. Let E+,E- be the linear spaces spanned by the e, with a # CCuna,and a > 0, respectively, a < 0; in other words by the e, with a > 0 and CC,, a < 0, respectively, with a < 0 and CCuna> 0. Then

CC,, E+ = E-. Put

E;=(1 +C)E-.

Ei=(l +C)E+, Then CE;

=E l ,

CE; = E l ; hence C,, E $ = E,.

61.2.1

+

By means of CC,,( 1 + C,,) (1 + C ) = ( 1 + Gun) (1 C) one shows the elementwise invariance of (1 + C,,) E: under CC,,, hence under C and C,,. Therefore

(1 + cun>E,+c G c . u n , and by 61.2.1 61.2.2

E;

= (1 + C,,) E:

Furthermore, GC.he = (1

+ EZ = G,,,, + E:. + C ) ( E ++ E - + H )

- Cun)(l

because the e, with a = CC,, a vanish under application of (1 - C,,) (1 + C ) . G C . h c = (1

- Cun)(E:

by 61.2.1

E:

E;

+ E , + HC)

+ &.he,

336

51-62.

REALITY IN LIE GROUPS AND ALGEBRAS

by 61.2.2 C

Gc,un

+ E &+

Now put

X = EZ

+

H~,hc*

HC,he.

Then Xis a solvable Lie algebra sub Gc, and Gc = GC.,"

hence,

+ G c . h e = GC,," + x = Gc,

Gc = Gc,,, + X.

No element # 0 of X is CC,,-invariant; hence, Gc,un n X = (01,

which proves the splitting for algebras. 61.3 The argument is extended to the groups by means of a general proposition : Proposition Suppose that G E Gru Top Con, A compact sub G, Bconnected sub G, U,, UB1-neighborhoods in A , respectively, B such that 1 is an interior point of U, UB. Then A B = G and GIB is compact.

Proof The compactness of A guarantees the existence of a I-neighborhood V, in B such that u-' V B a cU, UB

for all

Q E

A.

Then V BA C

A UA U, C AB,

V B V BA

C V BA B C

AB,

and so on. The connected group B is generated by Vs; hence BA c AB.

Now A(AB) = AB, (AB)(AB)c AB,

B(AB) = (BA)B C AB, (AB)-'

= BA C

AB.

Therefore AB is an open subgroup of the connected group G, hence A B equals G. Moreover, G / B is compact as a continuous image of compact A . 61.4 By 61.2 Proposition 61.3 applies to G = Gc, A = Gc,,,, B generated by X, which is solvable. The splitting is unique, since ux = 1 (u E Gc,,,, x E X)

62.

THE FUNDAMENTALGROUPS OF THE REAL TYPES

337

implies that on a suitable basis x is triangular as well as unitary, thus diagonal, whereas its eigenvalues are the exponentials of those of an element of thus positive, whence x = I . It follows from 18.1.5 that exp is in fact a homeomorphism of X onto X and that X i s closed in Gc. (See 4.13.) 61.5. Historical Note The preceding theorem is not found in the places to which it is usually ascribed. It probably originated in a paper by K. Iwasawa, Ann. Math. 50, 507-558 (1949) (Lemma 3.1 I).

62. THE FUNDAMENTAL GROUPS OF T H E REAL TYPES 62.1 Before being used in the computation of the fundamental groups of real semisimple groups, the terminology on the operations with fundamental groups of compact groups has to be slightly extended. Consider a G E Alg Lie Com (dim G < m ) ,

G = G“’

+ G‘2’,

a direct sum,

where G“’ is semisimple and C‘2’is the center of G . In G(2’real subspaces Ci:’ and Cli) = iC6:’ are marked such that after waiving in G‘2’

a direct sum.

G ‘ 2 ’ G‘2’ ~ CI + G‘2’ un,

Furthermore, it is agreed that H‘2’ = (-3‘2’ H6;’ 7

~

(-3‘2’ *I

3

H’ :L

=

G “,‘.

If H ( ’ )is a trunk of G “ ) , then H = H “ ) + H ( 2 )is a trunk of G . For a unitary semimorphism C, of C “ ) ,one defines

G,,

=

G::) + Ct’n’,

H un = H “un’

+ H‘2’ un,

H,, = H::’ + H y ,

As noted in 41.1, the notion of weight with respect to the trunk H makes good sense for a linear representation f of G as long as G‘2’ is conducibly represented by .f. Only such linear representations .f of G are admitted such that the group generated by ,f(G,,n)is compact. Then the group generated by f(Cl’,’) is the I-component of the center of that generated by f(C,,), thus compact; hence G ‘ 2 )is surely conducibly represented. The restriction of a weight h o f f t o H ( 2 ’is even a linear representation of G t 2 ) ;exp maps the addition group of H!,:,)homeomorphically with a compact image. So A(2’ is imaginary-valued on Hit) and real-valued on Hi:’. Any real-valued linear function on HI:’ can be obtained this way by starting with a suitablef. In 46.3 A = Hi*,gwas defined as the addition group of integral elements of

338

51-62.

REALITY IN LIE GROUPS AND ALGEBRAS

H;, that is, of possible weights of linear representations of a semisimple complex Lie algebra. It is now called A(C("). The foregoing leads to the following definition : Definition A(G(z))is the addition group of linear functions of H C 2that ) are real-valued on H::). Furthermore

A(G) = A(G('))+ A(G(2)). (More correctly, the G%estriction of an element of A(G) is in A(G(J)).) Again A f ( G )means the addition group generated by the weights off. The fundamental group of the group generated byf(G,,) need no longer be finite, though it is still finitely generated and thus a direct sum of cyclic groups. The dual (46.2) of an infinite cyclic group is isomorphic to the addition group of imaginary numbers mod 27ri. Its isomorphism type is denoted by Z,. The dual of the fundamental group is a direct sum of groups Z, (rn = 0, 1,2, ...). One can easily verify an extension of Theorem 46.5.3: Theorem The dual of the fundamental group of the group generated by f(G,,,) is isomorphic with A(G) mod Af(G). The duality takes the same sense as in Theorems 46.3 and 46.5.3.

62.2 Gc E Alg Lie Rea SS;G is supposed to be the adjoint of a Lie algebra isomorphic with G, thus certainly Gc is centerfree. Problem Compute the fundamental group of Gc.

62.3. Proposition The fundamental group of Gc is the same as that of Gc,un*

Proof The proof follows from 51.5.10 and the fact that GC,hcis homeomorphic

with the vector space GC,hc. 62.4 Gc,,, need not be semisimple; according to the classification, it might have a nondiscrete center. For this reason the terminology for the computation of the fundamental group has been extended in 62.1. The notions of 62.1 are now applied to GC,un,Com = G,,,,

+ iGC,,,,

instead of G.

Gc,,, is faithfully represented as a subgroup of the adjoint of Gcbecause Gc is centerfree. Therefore the role off will be played by Y.EGc.."YxecCI.X.

62.

THE FUNDAMENTALGROUPS OF THE REAL TYPES

339

62.5 To use weights, G is assumed in ordered maximally compact dressing with respect to C and the trunk H . As in 51.14,

C = LAC,,. In addition to C one considers Co = AC,,. Clearly

G,, which means the subalgebra of A-invariant elements. As a trunk of GC,un.Com and Gco,un,com one uses (1 A ) H .Now, according to the definition off, the weights offare the rootforms of G on H restricted to in H * . ( 1 + A)H. As in 51.21, one rather interprets H:.un,com and Hco.un.com * Then one arrives at the following: Gco,un.com =

+

Proposition The dual of the fundamental group of G,,,, sense of Theorem 62.1 is isomorphic to 4Gc,un,com)

mod

(or G,) in the

A4Gco,un,com),

that is,

4Gc,un,,om) mod (1 + A)A-(G)* Here, of course, one again uses the fact that restriction to ( I A)His translated in symmetrization with respect to A. I t suffices to compute the fundamental group for simple G,. For the compact case it is known, in the twin case it is the same as in the complex case, thus as in the compact case; these cases are now disregarded. In 62.6-7 the fundamental group will be computed for the inner types and in 62.8 for the outer types, according to the classification in Sections 52-53. The results are expressed in terms of the fundamental weights of GA.

+

62.6 G E L j , inner type; hence A = 1. The fundamental group is isomorphic to

mod Am(G)* First the rootforms of Gc.un,com are expressed in terms of those of G. Two cases must be distinguished with respect to the subscript j (compare Section 52): A(G,,un.,om)

62.6.1 p j appears in no positive rootform of CC.,n.Com with an even coefficient. 62.6.2 p j appears with an even coefficient in the rootform Gc,u,,com, which is the lowest positive one with this property.

C cup, of

340

51-62.

REALITY IN LIE GROUPS A N D ALGEBRAS

In the case 62.6.1 Gc,un,comhas a one-dimensional direct summand; p j simply drops out as a rootform. In both cases a natural basis of W*(Gc,,n,com) is furnished by p; = pi for i # j , p; = 2 c,p, if existent, that is, in the case 62.6.1 there is no p;. In the case 62.6.1 a basis of A(Gc,,,,,,,) is thus furnished by

.;

= Ti,

where integralcoefficients are admitted at ni with i # j a n d real coefficients at rl. In the case 62.6.2 an integral basis of /1(GC,,,,,,,) is available. One can use which have to satisfy (cf. 51.21) the fundamental weights r;of Gc.un,com (n;,pJ

-

2 T( p-k ~P k )

("1

for i # k , for i = k .

The T' can be expressed in the rr by

In Table F the 7~ are given in terms of the p. To find generators of the fundamental group one must compute the n' mod the p. This is done for the different types in 62.7. Notation The -sign is used for congruence modp,,

.. ., pl.

Notation Z,(a) for m = 1, 2, 3, . . . indicates the type of the m-cyclic group with the generator a mod A-(G,); Z,(a) indicates the addition group of TU with real T mod 1;here a, but no nonintegral multiple of a, is congruent to 0.

62.7. The Duals of the Fundamental Groups of Noncompact Centerfreet Simple Groups of Inner Type

-

Al,j: No p;. 7;= ri(i # j ) with integral coefficients, T; = rrj with real coefficients. Since T k k r , , one can take and v j as generators. For 1 cj < 1 + 1 put d= greatest common divisor of 1 1 andj,

+

d = u j + w ( l + 1). Then

dnI = ujv1+ w(l+ 1)

-

urj.

Though being centerfree is implied by simplicity, it is explicitly mentioned because Lie theory suggests a wider notion of simplicity, namely local simplicity.

62.

34 1

THE FUNDAMENTAL GROUPS OF THE REAL TYPES

Put 1’ $ = T I --7rj.

Then

d9.

-

d

0.

Touse Q , ras j a basis onemust show that no smaller integralmultipleofikan be congruent with a real multiple of 7 r j in, other words, that 62.7.1

x7r1

VI

y7rj

can be solved with integral x only by multiples of d. Comparing the coefficients of p I , p j , p j + , in 62.7.1, one gets congruences mod I + 1 : X I = y(1

+ 1 -j),

+ 1) = y(l - j + I ) j , ?I = y j m o d 1 + 1 ; if substituted x(l - j

x(l - j )

=

y(l - j ) j .

The last two give in the first, this gives y(l+ l ) j = y(1+ 1) mod 1+ 1, and since y j (=x mod I + 1) is integral, y ( f + 1) = 0 mod I 1, hence y integral. Therefore mod I 1, x is an integral multiple of j , thus d is a divisor of x. The result is

+

Forj = 1 this holds as well.

+

342

51-62. REALITY IN LIE GROUPS AND ALGEBRAS

B i , 2 : No p i .

Comparing the coefficients of pz and p3 (for 1= 2 those of pI and p z ) in TI =

f ( h+ p2 + 2p3 + . + (1 - 1) pi) *

*

and 772

=PI

+ p2 +

* * *

+PI,

one verifies that 7rI is congruent with no real multiple of 7rz. Z2(",)

+ Z(7r2).

7rl = p1

+ 2p2 +

771 = PI

+ p2 +

Cl,l: No pi. Comparing * * *

and * *

+ (1 - 1)pi-1 + g p i ,

-+

PI-1

+ 3p1,

one notes that a multiple of 7r1 cannot be congruent with 7rI unless I is odd. Zo(27rI) for odd 1,

D i , j ,j > 3 : p; =pj-l 7ri.2

+ 2p, +

= 771.2 - h

7r;=ni

* * .

+

Z2(7r1) Zo(7rl) for even 1.

+ 2p1 + p1 + p2.

j ,

for 3 < i < j - l ,

= 7rj-1 - +T,,

n; = h

j ,

rr; = r 1- 7 r j

for i>j.

may serve as generators; 7 r j belongs to the group generated by but $7r, does not. Indeed, if +rj YI u7rl u7r2 were true for some integral u,u, then the coefficieni, of p I , p 2 would yield u = vmod2, whereas the coefficient of p 3 would require u u = 1 mod2. The order of +7rj is 2 for even j , 4 for odd j .

7r1,7r2,+7r,

+

7r1,7r2

+

62.

THE FUNDAMENTAL GROUPS OF THE REAL TYPES

1 even : Z2(x1)+ Z2(n2)+ Z2(+7rj) z2(x4)

I odd :

-

Z,(nl)

Note that 7r2

for even j ,

+ z4(+xj)

for oddj,

+

for even j ,

Z4(~1) Z z ( + ~ j )

+ Z2(771-

-xl

hj)

343

for odd j .

for odd 1and x2 * xI+ irj for even 1 and odd j .

Dl,3: NO P;. 773 =

+(PI

+ P2) + P3 + P4 + . . + PI.

For even I: x , + x2 - n 3 ; for odd I: 2n, * x 3 . In either case xIis congruent with no real multiple of 7r3. 1 even:

+

Z2(nl) Z0(2~3),

1 odd : Z2(x1- 3x3)

+ Zo(2x3).

Note that x 2 * -xl for odd 1. D1,,: No p i .

For even I: 7r2 is congruent with no real multiple of xIas is seen by comparing the coefficients.

+

1 even : Z2(x2) Z0(2nl), 1 odd : zo(47rl).

5 1-62. REALITY IN LIE GROUPS AND ALGEBRAS

344

Generators 7rl, 47r2of orders 3, 2. z6(7r,

Note that 7r3

-

+ iT2)* -T,.

62. THE FUNDAMENTAL GROUPS OF THE REAL TYPES : P;

E8.2

F4.2:

=2P2

345

+ P3 + 2P4 + 2P, + 3P6 + 3p, + 4/38.

P i = 2 P l + 2 P 2 + 4P3 + 3P4,

(Pi9 P i )

= (P29 P 2 ) .

n;= 711 - +nz,

n; = fn,

+0,

n; = 773 - 7r2, 77; = n4

-

3r2.

Z2(3772>. G2.2

: p; X;

=

3 ~+ 12

~ 2 ,

(PA P;) = (PZ, ~ 2 ) .

= n, - 3772,

7r; = 3 n 2

+o.

Z2(+7r*).

62.8. The Duals of the Fundamental Groups of Centerfreet Simple Groups of Outer Type It is a matter of convenience to write the

symmetrized rootforms as p ; = +(Pi

+ 44.

The primitive rootforms of Gc,un,comare indicated by p;. The fundamental weights 7r’ corresponding to the p’ are those of Gco,un,com. They happen to coincide with the symmetrized 7rl except in A 2 m , 0 , * , where +(n, i- PX,) = 2n;;

this is precisely the case in which not all symmetrized rootforms are rootforms (See 53.1.) of Gco.un.com. Cf. footnote in 62.7.

5 1-62.

346

REALITY IN LIE GROUPS AND ALGEBRAS

Forj # 0 one needs the fundamental weightsdcorrespondingto thep", which are derived from TT'by the method used in 62.6 to derive d from n. Now v, denotes congruent mod the p'.

+

A2m-l.0,*: p; = +(pi pzm-i).The fundamental group is the same as in C,. ZZ(4).

A2m-l.m,*: p;=p; n

(i=l, t

I

I

Pm=Pm-l

+p:,.

( i < m - I),

r i= r i 4 - 1 =d - 1

..., rn-11,

-

MI,

7r; = +n;.

+ +

+

+

Now 47; = +(p' 2p; * * (m- l ) ~ k - ~trnp;) comes up, whereas in C, one could do it with T ;alone; i n ; $. n;.

For even m : Z2(7r;) + Z2(4r;). For odd m : Z4(+m;). DL~.*:p;=pi

for i # l , 2 ,

p;=+(pI+p2).AsinBi-l,o.

Z2(773

In addition to 7 ;one gets +T;as a generator. Comparing the coefficients of p; and pi in

+ p; + 2p; + = + ( ( j - 2)p; + p; + 2p; +

7r; = +7r;

$((I

one verifies rr;

- 1)p;

+

+r;.

Z 2 ( 4 ) + Zl(f-7;).

* * *

''*

+ (I- 2)p;), + ( j - 2)(p; +

* * *

+pi)),

62.

THE FUNDAMENTAL GROUPS OF THE REAL TYPES

ir; = ir; -

347

+n;,

62.9. The Duals of the Fundamental Groups of the Universal Linear Wrappings In the compact case any wrapping (being compact again) can be realized as a linear group (see 46.8). This is true of the twin type as well, but in many other real types such a linear realization need not exist. The linear wrappings of centerfree G, are real restrictions of the linear wrappings of G and are the I-components of the originals of G, under the projection mappings. It follows that the 1-component of the original of G, in the universal wrapping of G is the universal linear wrapping of G,. The splitting of G, into connected G,,,, and simply connected GCSheis lifted to such a splitting of any wrapping of G,. Hence the fundamental group of the universal linear wrapping of G, is the same as that of the induced wrapping of G,,,,. The addition group generated by the weights of the complexification of this representation of G,,,, is that generated by the fundamental weights of G, symmetrized in the outer case. Hence by 62.1 : Theorem The dual of the fundamental group of the universal linear wrapping of G, is A(G,-,,n,co,,,)mod the group generated by the fundamental weights of G, symmetrized in the outer case. The particular cases are:

G, compact or twin: Trivial.

348

51-62.

REALITY IN LIE GROUPS AND ALGEBRAS

62.10. Historical Note E. Cartan found the fundamental groups of centerfree semisimple Lie groups by rather casual methods [Ann. Ecole Norm. 44, 345-467 (1927) = Q3uures I 2, 867-9901. Current folklore says that E. Cartan dealt with the compact case only, though the authors who claim to have solved the noncompact case first and others who are supporting this claim do not fail to quote the paper in which Cartan published his results. Since there is some danger that modern research will eclipse Cartan's, it is useful to add the warning that those modern results are partly mistaken, even on primordial points, whereas Cartan's are completely correct.

63-67 SYMMETRIC SPACES

63. HOMOGENEOUS SPACES AND RIEMANNIAN MANIFOLDS-A SKETCH 63.1-3.

Homogeneous Spaces

63.1.l.Definition The pair ‘G,J1, in which G is a topological group and J a closed subgroup of G, determines the homogeneous space

R = G/J that is, the left coset space ofJin G, which is acted on by G by left multiplication in a transitive manner. For any group G acting on a space R the stability group o f p E R consists of the elements of C leavingp invariant. 63.1.2 For the just-defined homogeneous space the stability group of xo = J is J itself; generally that of a point g, xo (gl E G) is g,Jg;’.

63.1.3 How doesJact on R ? By left multiplication b y j E J , xJis mapped into

jxJ =jxj-l J so that J acts on R as a subgroup of the group of inner automorphisms of G. 63.1.4 It goes without saying how the homogeneous space G/Jis to be understood as a direct product of homogeneous spaces G,/J, if G, respectively, J , is the direct product of G,, respectively, 5,. 63.1.5 The representation of G in R = G/Jis faithful iff Jcontains no normal subgroup # { l } of G. Generally this will be assumed in the sequel. 63.1.6 G I sub G acts transitively on G/J iff GIJ = G; in other words, if every left coset of J intersects GI or, equivalently, if every right coset of G I intersects J. Then Gl/GI n J is one-to-one continuously mapped onto G/J by assigning a J to a (GI n J ) for all a E GI. This mapping becomes homeomorphic as soon as G and G I are locally compact, second countability Hausdorff spaces such that 4.8.4 applies. 349

350

63-67. SYMMETRIC SPACES

63.1.7 A homogeneous space G/J is called reduced if there is no G I sub G acting transitively on G/J except G itself. 63.1.8 G may have several components. Its 1-component will usually be denoted by Go.In the examples to be considered Go is isolated, that is, open, in G. Then Go/Gon J is open in GIGo n J, and if Go/Gon J is mapped into G/J by mapping a (Go n J) (a E Go)into aJ, then the image is open and connected, hence equals Go/J. Applying 63.1.6 and remembering that the cosets of Go in G are the components of G, one gets the following: R = G/J is connected iff every component of G has a nonvoid intersection with J o r , in other words, iff every left coset ofJhas a nonvoid intersection with GO.

63.1.9 A homomorphism of G onto G’ which takes Jinto a closed subgroup J ’ of G’ and is locally topological on G extends to the homogeneous spaces GIJand G’IJ‘ as a locally topological mapping. On the other hand, leaving G unchanged, but replacing J by a larger, closed subgroup J ‘ in which J is open, induces a locally topological mapping of G/J onto G’IJ’. These remarks may be summarized by stating that the local structure of R depends on the local structures of G and J only. 63.2 In the sequel attention must be paid to groups locally isomorphic to linear Lie groups. Their components are Can-manifoldsand their l-components are Can-groups(see 6.4). Let G , be such a group and x the given local isomorphism from the linear Lie group G to G I . If no confusion is possible, in all local investigations near 1 G and GI will simply be identified. Among others, the infinitesimal algebra G of G will also be called infinitesimal algebra of GI. Any a E G induces an automorphism in G, to be denoted by 5, since the inner automorphism induced by a near 1 is analytic by Theorem 1 1.4. 63.3.1 Let G be a group locally isomorphic to a linear Lie group, J a closed subgroup, G, J the corresponding Lie algebras, and N a linear subspace of G such that G =J + N, a direct sum.

If U is a small 0-neighborhood in N,then exp U intersects x J for x near 1 in exactly one point. The mapping rpa, which assigns to XJfor x near a in G the only member of a-’xJ n exp U,is an analytic mapping of an a-neighborhood in R = G/J onto a 1-neighborhood in exp U.In the sequel R is supposed to be connected. Then with the rpa as local presentations of R (see 5.6), R becomes an analytic manifold on which G acts as an analytic (nonlinear) Lie group.

63.

351

HOMOGENEOUS SPACES, RIEMANNIAN MANIFOLDS

Near xo = J, respectively 1, R can be identified with exp U by short, tp.

tpl

or, for

63.3.2 The behavior o f J a t xoand, more particularly, in the tangent space at x,,, is worth studying. Of course, Jwill act in the tangent space as a linear group.

The inner automorphisms G of G as acting on Gform a linear group acting on T = 0 of the curve y, exp in exp U. N o w j ( j E J) maps exp T I I for small T into G. Take n E U. Then n is the tangent vector for

nnl(r)j1(~), j e x p = exp ~ j = wheren,(~)E exp U,j1(7)E J,nl andj, beinganalyticfunctions. Differentiation at T = 0 shows

j n = n ] +i],

where

Relating exp U to an xo-neighborhood in R by one may consider nl(.) as thej-image of exp in R, and nI as thej-image of n in the tangent space. It has been shown that nI = j n mod J. Therefore the action of the stability group J in the tangent space of R at xo is described by that of the representation of J as a subgroup of Int G, induced in G mod J. Note, however, that this description of J need not be exhaustive. Nontrivial j ( j E J) may be trivial on the tangent space. An example: the projective group on the straight line-the translations, that is, the projectivities with a double fixpoint at infinity, induce the identity mapping in the tangent space at infinity. In such cases the behavior of J can be fully described with tangent spaces of higher differential order.

63.3.3 Suppose now that N in 63.3.1 can be chosen to be invariant under the i ( j E J).This is possible when J is compact or semisimple and connected; then the representation of J in G as a subgroup of Int G is conducible, and J, being invariant under j ( j E J),has a linear complement that is also invariant.

In this case the behavior o f j in R is fully described by that ofjin Nbecausej, considered as acting on R, is determined by its action on any nonvoid open set of R (which, with its inverse, is analytic). The representation of J in N by y , j is faithful because it is so on R. Identifying R locally with N, one may say that J behaves as a linear group near xo.

352

63-67.

SYMMETRIC SPACES

63.3.4 In terms of fiber spaces, exp U,as introduced in 63.3.2, is a local cross section near 1 of the fibering of G by theJ-cosets, as is U,, = a exp U near a for every a E G. These cross sections can be used to relate the fundamental groups @ ( J ) , @(G), and @ ( R ) . Theorem Let C be connected and locally isomorphic to a linear Lie group, let J be connected closed sub G, l e t j be the embedding o f J in G, and k = 'fad. Then the sequence of (additively written) fundamental groups

@(J)

: @(G)

k

+ @(R) + (0)

is exact. Proof Let U,, be as before. Then klua is homeomorphic and its inverse lifts a

neighborhood of ka onto U,,. It can first be used for piecewise and then for total lifting of any given path I V E wka(R) into a path id E wa(G). Then kid = itJ. Let )(lrkl(R) and yf1(G) be topologized in an obvious way; k (more precisely left multiplication by k ) maps yfl(G) continuously onto w k l ( R ) . For any iv E 9fl(G) let V , be the set of w' E "w;(G) with "'(7) E Uw(r) for all T . Then kl,, is homeomorphic, and its inverse lifts a neighborhood of kw onto V,. Again any path in w k l ( R ) starting at kiv can be lifted into a path in wl(G) starting at IV. To return to the statements of the theorem, it is evident that for the fundamental groups kj is the null-homomorphism. One must still prove first that k is onto and second, that the kernel of k is contained in @ ( J ) . Let a closed path iv E W l l ( R )be given. It is lifted intoa path w' E "Wl(G), which necessarily finishes in J . Since J is connected, id can be lengthened by a path withinJto get a closed path w" E Wll(G)such that kw" IS homotopic with the given w. This proves the first part. Now let iv; E 7Y11(G)be given such that kwh is homotopically trivial. This means the existence of a path )'owo on 'W:f(R) such that w o = kw; and I V ~ ( T= ) kl for all T . This path is lifted into a path Y,wA on Wl(G).Then W ; ( T ) , iisA(1) E J for all T , U . By 28.11 w; is homotopic with w ; o 'f,,~vA(l), which lies in J. This proves the second part. 63.4-6. Geodesics 63.4. Definition A differential metric on a real C2-manifold R assigns to the tangent space of R at any point p of R an inner product, depending twice continuously differentiably on p . The inner product is supposed to possess the usual properties, positive definiteness included.

63.

HOMOGENEOUS SPACES, RIEMANNIANMANIFOLDS

353

A C2-manifold with a differential metric is called a Riemannian space. By integration the notion of vector length in the tangent spaces extends to a notion of curve length for C1-curves, invariant under weakly monotonic parameter change. After such a change any C ‘-curve may be supposed to have a vanishing tangent vector at its beginning and end. Two such curves, the second starting where the first ends, can be put together to form a C1-curve, with additivity of the curve length. The distance S ( p , q ) of two points p , q of R is defined as the lower bound of the lengths of C1-curves joining p and q. Obviously it is nonnegative symmetric, and from the foregoing remarks it follows that it fulfills the triangle inequality. It will soon become clear that even 6 fulfillsthe positiveness requirement of a metric and that the topology induced in R by the metric 6 coincides with the original topology of R as a manifold.

Definition A geodesic p is a C1-mapping of an interval I of real numbers into R with the following property: for any T E I there is an E > 0 such that T’ E I and IT - 7’1 < E imply S ( ~ ( T ) p , ( ~ ’ )=) IT - 7’1. Geodesics arising from a given one by a monotonic parameter change or by extending or restricting the interval of definition are often considered essentially equal. The classical way of finding geodesics is to look for shortest curves. Indeed, compared with other C’-curves, any partial curve of a shortest curve is still a shortest curve between its endpoints, and, as will be seen, any shortest curve arises from a geodesic by a suitable reparametrization. On the other hand, it is evident that a sufficiently small arc of geodesic around a given parameter value is a shortest curve. The variational approach to shortest curves will be sketched briefly. The differential metric is transferred from an open neighborhood Ro ofp, in R to a 0-neighborhood Eo in real Cartesian n-space E by a C2-homeomorphism f with nondegenerate gradient and fpo = 0. The discussion takes place in Eo. At a point x of E, the inner product of two tangent vectors

is given by

where rg,,(x)ly,,=I is a positive definite symmetric matrix, depending twice continuously differentiably on x . The length of a C’-curve p in Eo from 0 to y is given by

63-67.

354

SYMMETRIC SPACES

63.4.1

(Generally q' now denotes the derivative of 9.) The curve length is invariant under weakly monotonic C1-changes of the parameter of q ~ . To compare curve lengths, a family of C2-curves v0 is studied; it is defined by a C2-mapping @ of a rectangle t rT,,,l (0 Q T Q CL A 1u1 G /3) of the Cartesian 2-space into Eosuch that @(O, 0 ) = 0, 63.4.2

( c ~ ' ( T ) ,c ~ ' ( T ) ) ~= ( ~1)

for 0 < Q a,

where it is understood that

v'

= Y T @(',

lp = TO.

Furthermore 8 is defined by 63.4.3

and the coordinates of the introduced functions are indicated by subscripts in the usual way. An easy calculation that involves a partial integration leads to the formula

whereg,J,kis the partial derivative ofg,, in the kth coordinate and where, under the integration sign, the arguments T , respectively, (p(~), have to be read in (pi,(p;,8k? respectively, gik, g I J . k The coefticient o t ~ under k the integration sign can also be written as 63.4.5

Dk(T) =

-7

gik

v; f

IJ

(gIJ,k - gik.i - gkI.J)

v;,

where lp: is the second derivative of (p,. To find a necessary condition on )'04TGrq(~) being a shortest curve, a family of curves with the same endpoint is chosen; that is, @ ( t , u) independent of u for some fixed t .

Then 8(t)= 0, which causes the first summand in 63.4.4 to vanish.

63.

HOMOGENEOUS SPACES, RIEMANNIAN MANIFOLDS

355

For the length of ‘fOcsBty.0()to be minimal, one must have

By suitable choices of that family 8 k can be prescribed with such a degree of freedom that, as a necessary condition for rp to be a shortest curve, one gets 63.4.6

&(v)

for all k.

=0

This is now considered as a second-order system of differential equations to be solved with respect to cp under the initial condition

940) = 0, T’(0) = a, where a is a given vector. Its (unique) solution is va.Clearly ~ f i a ( 7 )= r ~ a ( 8 7 )

for real 8 as far as the expressions make sense; the qnwill turn out to be shortest curves. One can easily verify that

because of the differential equations 63.4.6 (see also 63.4.5); hence (VX719

V:(.Dcp.(.r)

= (a, a>,.

Therefore, if (a, a)o = 1, then 63.4.7

Lt(qa) = t .

If Eo has been chosen small enough, it will be smoothly covered by the solutions vaof 63.4.6 in the following sense, Near 0 0

=

Y

(t

Va(1)

is Cz with gradoa = 1, hence Cz-homeomorphic near 0. Since, for a # 0, Ya‘lal,

IaI-la’

is analytically invertible, it follows that for Eo small enough and x E E,\{O} there is one positive number S(x) and one vector c(x) with (c(x),c(x)),, such that 63.4.8

VC(Z)(W)

=

1

=x

and ‘S,cl is a Cz-homeomorphic mapping with a nondegenerate gradient. In the sequel Eo is assumed to fulfill this requirement.

356

63-67.

SYMMETRICSPACES

Note that tx(six) = p ) may then be considered a C2-manifold (for p > 0). The previously introduced family of curves is now specialized to consist of curves y,, : 63.4.9

@(T,

u) = vrr(,,)(7)

where (a(o), U ( U ) ) ~= 1 for all Then 63.4.10

CT

1.(

for 0 G 7 G t ,

G some

/3) and a(u) is a C2-function of u.

Mqa(oJ =

The derivative with respect to u vanishes, and in 63.4.4the second summand on the right side vanishes as well, since ya fulfills 63.4.6 if substituted for y . This shows that

= 1 meets the surface tx(8(x) = I ) This means that every curve vnwith at ?,(t) orthogonally in the sense of the inner product (. .., . . .). The time has come to compare an arbitrary C1-curve q in Eo, starting at 0, with the curves yo.Such a curve 7 may be assumed in the form

7(7) = 9)a(7)(47))

for

7

> 0,

where the a(.) are vectors with (u(T), ~ ( 7 ) = ) ~I , E(T) 2 0, and ‘u(T), ~ ( 7 ) ’is a Cl-function Of T. Since shortest curves are wanted, one may even suppose that E(T)

>O

for T > O .

Now

According to 63.4.11, the two summands are orthogonal to each other in the sense of the inner product (. .., .. .) at q(t). Hence

with the equality sign only if a’(r) = 0.

63.

HOMOGENEOUS SPACES, RIEMANNIAN MANIFOLDS

357

Integrating this inequality, one gets

with equality signs only if -q is such that U ( T ) is constant and E is weakly monotonic, that is, if 7 has arisen from some cpo by a weakly monotonic parameter change. This shows that for CI-curves T with ~ ( 0 = ) 0, q ( t ) = cpa(t), where (a,a),

=

63.4.12

1:

inf, L,(q) = t = Lt(cpa).

The C’-curves used in the definition of distance on R were allowed to leave the neighborhood R , ofp,. It is easily seen that this does not lower the infimum. Then by 63.4.12 it follows that thef-originals of the cpa are geodesics on R. From the remark on the equality signs it follows that any shortest CI-curve in E, starting at 0 coincides with some reparametrized cpp. The notation cpo, or, more completely, ,c.p, is now used on R to indicate a geodesic starting at p, with the tangent vector a of unit length. The result is summarized in the following: 63.4.13. Proposition Geodesics are C2-curves. Givenapointp of R and a vector a of unit length in the tangent space of R at p , there is an essentially unique geodesic c,p, with cp,.(O) = p , &JO) = a. There is an E, > 0 such that there is a unique geodesic starting from p and finishing at x of length 6 ( p , x )as long as 6 ( p , x ) < E,. On a compact set of p such an E, may be chosen as a constant. Then cp,,(~) depends on the datap, a, and T in a C2-fashion. From this it is clear that the topology of R as a manifold coincides with that induced by the metric. 63.4.14. Proposition The mapping u = Yaqp.a(l)

of a sufficiently small 0-neighborhood of the tangent space at p into R is a C2-homeomorphism whose gradient at 0 is the identity. This mapping is closely akin to the exponential mapping of infinitesimal Lie algebras into their groups. Its inverse is called the geodesic mapping (at p ) . It can be used to identify near p the Riemannian space with its tangent space at p in a canonical way. 63.4.15 If, for a moment, u is used to identify R nearp with its tangent space, the geodesics through p (=O) are rectilinear. The C2-data of the differential equation for geodesics lead to a C2-family of solutions. Let the geodesic of

358

63-67.

SYMMETRIC SPACES

length IcI starting at a with a tangent vector c/lcl have its endpoint at 6(a,c); then 6 is C2,a(0,c) = c, 6(a,O) = a ; hence #(a, c ) = a

+ c + ., * *

where the dots mean something of an order higher than la1 a’ = 6(u,c) and solving with respect to c, one gets c=a’-a+

+ IcI. Putting

*..,

where the dots mean something of an order higher than la1 + la’[. Since IcI = S(a,a’), one gets if lim a = lim a’ = 0. Therefore up to higher order terms the Riemannian metric in Rand the metric induced by the inner product in the tangent space at p are identified by (T. 63.5.1 If a metric space ‘R,Sl is known to have arisen from an unknown differential metric, as defined in the beginning, the C2-structure of R and its differential metric are (uniquely) recovered as follows : Forp E R choose y > 0 such that unique geodesicjoining prevails for pairs of points in the closed y-ball U around p. In U the midpoint z of a pair x , y (i.e., S(x, z) = S(y,z) = S(x, y)) is uniquely determined by 6. Given q E R with S(p, q) = y, define a continuous mapping Qrp of t7(0 G T Q y ) into R by the requirements “ ~ ( 0=) p, %p(y)= q ; ‘+p($(~, T J ) = midpoint of %p(~,)and % p ( ~(its ~ ) existence is clear). The Y,“rp(aT) (as far as defined), with q E R, S(p, q) =y , a 2 0, form a set Q p . The meaning of Q p becomes clear if one maps a E OPinto its tangent vector at T = 0, denoted by KU, according to the C2structure from which ‘R,S1 has been derived. The existence of the metric S’in QD, defined by S’(a,b) = lim T-’ S(U(T), b ( ~ ) ) ,

+

+

7-0

then becomes clear, as does its identification by K with the metric in the tangent space atp, from which ‘R,6’ has arisen (see 63.4.15). In a linear space with a metric derived from an inner product the linear structure is determined by the metric and the origin, as is the inner product. In this way djPis uniquely provided with a structure of linear space with inner product, which by K is identified with that in the tangent space a t p of ‘R,S1. Finally the C2-structure of R is recovered by yvE@,,rp(l) (as far as defined) as a local presentation of R near p ; the differential metric at p is recovered as the metric of aP. The foregoing leads to a proposition and a remark:

63.

HOMOGENEOUS SPACES, RIEMANNIAN MANIFOLDS

359

Proposition The differential metric of a Riemannian space is uniquely determined by its metric. Remark In the definition of geodesic "C'-mapping" can be weakened to

"continuous mapping." 63.5.2 The direct product of Riemannian spaces R") ( i = 1,2) is easily defined as the direct product of the underlying C2-manifolds, where the tangent space at rp(l), p(2)1and the inner product in it are the direct sum of those at p") E R") ( i = 1,2). On the other hand, a metric product ' R , 6 l of the resulting metric spaces rR(i), ( i = 1,2) is formed by putting for p ( ' ) ,q") E R")

a( r p ( l ) , p ( 2 ) 1 ,

rq(l),

pi)= ( g c l ) ( p ( l ) , q ( 1 ) ) 2

+ g(2)(p(2),4 2 ) ) 2 ) 1 / 2 .

As a matter of fact, ' R , 6 l is just the metric space resulting from the direct product of the R"). Indeed, this easily follows from the differential equation for shortest lines in R . It splits in those with respect to the R") such that if v") is a solution for A(') (i = 1,2), then y t rq+l)(r), q ~ ( ~ ) (ist )one l for R. 63.6 An important tool in global research on Riemannian spaces is the postulate of Geodesic Latitude Any geodesic can be extended to one defined on the whole set of reals. This property will now be assumed. Proposition Under geodesic latitude, any pairp, q of points of R can be joined by a geodesic of length 6 ( p , q), and any bounded closed subset of R is compact. Proof Let R , be the set of x 6 ( p , x ) from p to x.

E

R such that there is a geodesic of length

The first step is to show the following: 63.6.1 R, is closed and every bounded closed subset of R, is compact or, equivalently, 63.6.2 a bounded sequence x,, x2,

in R,.

... from R, accumulates somewhere

360

63-67. SYMMETRIC SPACES

Such a sequence may be supposed to give the yi= 6(p,x,) a limit y. There is a geodesic visuch that ~ ~ (=0p,)vi(yi)= x l . Now the sequence of ( ( d / d ~rpi(7)>0 ) may also be supposed to converge. The limit is the tangent vector at 0 of a geodesic rp with rp(0) = p. By geodesic latitude the geodesic rp may be extended for all T > 0. Proposition 63.4.13 implies that lim xi = lirn v i ( y i )= rp(y) = X, say. Further, 6(p,x) = lim S(p,x,) = lirn y i = y. This proves the assertion lim x1= x E R,. The next step is to show that

R,

63.6.3

= R.

Let S, be the set of x with 6(p, x) Q 0: and let y be maximal such that S, c R,. If R, # R, then y < co. Sy is a closed bounded subset of R, and thus compact. According to Proposition 63.4.13, let E > 0 be chosen such that for z E S, and 6(z,x) G E there is a geodesic of length S(z,x) from z to x. If R,, # R, then there is an x $ R, with S(p,x) < y + +E. Let q~ be a curve of length < y E that joinsp to x. It leaves Sy at some pointy. Its length fromp toy is> y ; therefore, fromytoxit is< E. Let z E S, beclosest tox. Still 6(x,z)AI) = ( A 3- [ A ,AD n W 2 + [ A ,4) = A n N2 = (0). Thus J , is an ideal of G contained in J. Hence

J1 = { O ) ,

J2= J,

Since

[J, [N,Nll = “J, NI, NI

Nl = A .

+ [N, [J,NIL

[ N , N ]spans an ideal J’of J. Therefore J’ + Nis a subalgebra of G which has to coincide with G, since N is supposed to generate G. Consequently, [ N , N ] spans J. This shows that J # (0) unless G is abelian. Put GI = J + N2. Then G1is a T-invariant subalgebra. Since

[ A , N 2 ] = A n[ N , N ] = A n J = { O ) , and [A, J ] has the same span as [A,

9 “

Nl1= [A, “2, N211c “A,N2I, N*1+ “ 2 , [A, N211 = {O),

one gets [ A , Gll = (0); thus

+

G = GI A ,

a direct sum of algebras.

G1is semisimple, since otherwise the last but one member of the commutator sequence of its radical could be added to A in spite of its maximality. Being semisimple, G1 may possess a proper simple direct summand G ; . Then G; TG;equals GIor is a direct summand of GI. This procedure can be continued.

+

Definition A solution rG,T1 with G # (0) of Problem 64.4 is called sirnp!e if G has no nontrivial proper T-invariant ideal. The following has been proved: Theorem Any solution of Problem 64.4 splits directly into simple solutions, some semisimple and some abelian. For an abelian solution J = (0).

369

64. SYMMETRIC SPACES

64.6 The investigation now turns to the semisimple simple solutions.

Let ‘G,T 1 be semisimple simple and let J, N be as before. There are then three possibilities:

Theorem

(1) G,T have arisen from complex semisimple G with an involutory automorphism by waiving. (2) Under the representation of J in G as a subalgebra of 6 the subspace N is irreducible over Com and J i s semisimple. (3) Instead, N,,, splits into two nonequivalent irreducible linear subspaces, both of which are abelian subalgebras; the group infinitesimally generated by adc J J Ncontains the 1-component of its centralizer within the group of all volume-preserving linear mappings of N , and if N i s irreducible over the reals, this centralizer itself; and J is the direct sum of a semisimple and a onedimensional Lie algebra.

Complexify G (with the conjugation Co), extend T to G,,, automorphism such that CoT = TC,, and suppose that

Proof

NCom =

N , + N27

NI n N2 = {O},

[J,Nt] C

as an

Ni*

If n, E N,, n; E N,, then by Jacobi-associativity “n,,n;I,n,l+ “n;,n2l,n,l+ “n,,n,l,n;l

=o,

where the first summand is in N2 and the others are in N , , which shows “ N , , NIL N2l = (0) and likewise “N2, N21, N11 = {O}.

Let JI be the maximal subalgebra of Jcomwith [ J , ,N21 = {OI.

Then by the foregoing

“,,N,I=

JI.

Jacobi-associativity shows that J , is an ideal in Jcom.Let N ; be the linear space spanned by [J1,N,].Then J , N ; is a T-invariant ideal of G,,,. The same is true of its complex conjugate CoJl CON;.Their sum and intersection, if Co-restricted, must be G or (0). If the C,-restriction of the intersection is G, then J , + N ; = G,,, and N2 = (0). If it is (0) and J , N ; # {0},then the Corestriction of the sum is G, and Gmay be supposed to be obtained from JI N ; by waiving, which is the first kind of solution indicated in the assertion. Therefore, disregarding the first and second kinds, one may suppose that N2 # (0) and J , N ; = {0},hence J , = (01, [N,,N,I= (0).

+

+

+

+

+

63-67.

370

SYMMETRIC SPACES

This shows the following: 64.6.1 If N , # {0},N2# {0}, then [N,,N,]= {0}, and for each i the only element of J,,, commuting with all elements of N1is 0. Suppose N 1 , N 2as before; adc,,,N, is nilpotent: it maps N , into {0}, its square maps Jcominto {0},and its third power maps N2 into (0); of course, N2 is also ad-nilpotent in C,,,. t,b is nondegenerate on N,,,, since #(. Ncom) I, = (0). However, t,b(Ni,N,)= {0}, since N , is ad-nilpotent. Thus

dim N1= 3 dim N,,,.

64.6.2

Since every J,,,-invariant linear subspace of N,,, has an invariant linear complement, the following applies : 64.6.3 The N , are irreducibly acted on by Jcom. I t will now be shown that these actions cannot be equivalent. Equivalence would mean the existence of linear mappings 6, (i = 1, 2), respectively, of N , onto N2 and of N2 onto N , such that

64.6.4

j8,n, = 8,j n ,

for all j

64.6.4a

j62n, = 8,jn,

for all j E Jcom,

E

Jcom,

"1

EN,,

n2 E

N2.

By 55.4.4, if the Ni are real (i-e., C,-invariant), 6, and 6, may be taken to be real; if the N , are nonreal, they may be taken as each other's conjugate and so may 6,and 6,; hence 6 2 = Co8,

c,.

+

Then the linear subspace of the n l 8 , n , ( n , E N , ) is still invariant under j ( j E .Icom)because of 64.6.4. So it is again abelian; hence [n,

64.6.5

+ 8,n,, n; + 9, n ; ] 0 for n , , n ; E N , , [a,n,,n;1 + b,,4 4 1 = 0. =

Likewise

[6,n,, n;] + [n,,19~n;] = 0 for n2,n; E N,. Now define the linear mapping 6 of G,,, into itself as 0 on J,,, and coinciding

S4.6.5a

with 6,on N,. This 6 appears to be an infinitesimal automorphism of C,,, because of 64.6.441, rewritten as

W, nl = [ j ,94 +

4,

64.

SYMMETRICSPACES

371

and because of 64.6.5-5a. (Note that the other commutator relations are trivial.) Because of the reality requirements, 8 may even be considered as an infinitesimal automorphism of C . Now because G is semisimple, 8 is inner and 9 i for some u E C, but, since 8T = T8, u belongs to J. This is contradictory whereas 19 does not. Therefore the representbecause u' (u E J ) preserves Ni, on N , and N2 cannot be equivalent. ations of Jcom Finally the centralizer, as mentioned in the statement, consists of the restrictions u, to N of linear T , defined on N,,, by :

~ , n= , mini for a =

r u I , u 2 1with

for ni

(~,CL~)= ~ ' 1," "and ~ a l r u 2real

a I= u2

for real N I , N 2 ,

for complex conjugate N , , N2.

u, extends to an automorphism of G with u,j = j f o r j E J a s soon as a Iu2 1. In the case of conjugate Nithis is fulfilled by all u with T , in the centralizer, which is connected; for real NLit is still true for u with T , in the I-component of the centralizer. Therefore the I-component of the centralizer, as acting on G, is within Int C and, since 0, T = Tu,, even within the group infinitesimally generated by ad,& However, by restriction to N this group is homomorphically mapped onto the group infinitesimally generated by a d G J (N , and by this homomorphism 0, is mapped into urrlN. For the structure of J in the second part of the statement, that is, if J acts complex irreducibly on N, then up to a direct summand of scalar multiplications, spanned by some z , J i s semisimple. However, because o f t j = 0 for j E J, t n = un for n E Nand some u E Rea, one gets on the one hand Z[n, n'] = 2u[n, n'] for n, n' E N , and on the other [n,n'] E J ; thus t [ n , n'] = 0, whereas [n,n ' ] # 0 for some choice of n,n'; hence u = 0. Then z would be a center element of G belonging to J, which is not allowed. Therefore Jis semisimple. I n the third case of the statement it was proved that the center of J i s at least one-dimensional, and it follows from 64.6.1 that it is at most one-dimensional. This completes the proof of the theorem.

64.7 In 64.2 the point of departure was a symmetric space R . From this datum a group IntOR of autometries with an involutory automorphism T was derived. In 64.3 the procedure was reversed: Starting from a group G (say IntOR) with an involutory automorphism T (with certain properties), a symmetric space was constructed. To know all symmetric spaces it is useful to classify all Lie algebras G with such an involutory automorphism. A more general problem has been tickled in 64.4. As a first step, 64.5 made the restric-

372

63-67. SYMMETRIC SPACES

tion to simple semisimple solutions advisable. These solutions have been thoroughly analyzed in 64.6. To take up the thread where it was left after 64.3, one has to turn back to groups instead of Lie algebras and, in particular, urge that J,, infinitesimally generated by J, has a bounded adjoint action on N. Theorem 64.5 suggests the following: 64.7.1. Definition Let R €Spa Sym, C=IntoR, J b e the stability group of p E R in G, and let T be the automorphism of C determined by s., Then R is called abelian, semisimple, or simple,? depending on whether such an attribute applies to ‘G, T1. Proposition A symmetric space R is locally the product of simple symmetric spaces R,.

+

Proof It may be supposed that R is simply connected (see 64.1.3). G = J N as in 64.2. G splits directly into Gi ( i = 1, ..., rn) such that the rG,, TI are simple (see 64.5) and J, Nsplit correspondingly into J,, N,. The Gi, J, may be considered as generating G I ,.I,, which are locally isomorphic with linear Lie groups, Ji being closed in Ci. Homogeneous spaces R, are defined as GJJ, ; their

product is denoted by R*; Ri is considered as a subspace of R* in a natural way. A mapping h of R* is defined by

R* wraps R by means of h; since R is simply connected, h is homeomorphic; it is now used to identify R* with R 3s homogeneous spaces. N considered as a tangent space of the Riemannian space R = C / J is endowed with an inner product (. .., ...). The splitting may be refashioned so that the tangent spaces Ni of abelian R, at J, are orthogonal to each other. Then the tangent spaces N , of all R, at .Ii are orthogonal to one another. Indeed, let R, be nonabelian, thus simple semisimple; then one obtains fork # i : (NkrN J c (Nk, [J,, N , ] )= ([Ji, Nk],N , ) = (0) because of infinitesimal invariance and of [Ji,

Nkl

=

Therefore at J the tangent space of R is the direct sum of those of the R,, even with due regard to the inner product. This property is transferred by the action of G to any point of R. Thus the Riemannian space R is the direct product Riemannian subspaces R,. Geodesics in R, are also geodesic in R (see 63.5.2). Therefore, if s, is the

t It would be more appropriate to call these spaces locally simple and to reserve the term “simple” for those that do not split globally.

64. SYMMETRICSPACES

373

reflection of R in p E R,, then splR, is the reflection of R, in p. R, possesses a reflection in each of its points. The even products of the s, with p E R, form a closed subgroup (even a direct factor) (IntOR), of IntOR, which leaves R, (k # i ) pointwise invariant. By restriction to R, it is identified with IntOR, which in this way is locally isomorphic to a linear Lie group. This proves the R, to be (simple) symmetric spaces, the product of which is isometric to R. Remark The proof shows that in a splitting of a symmetric space as a

Riemannian space, the factors are again symmetric. A simple symmetric space is also “simple” as a Riemannian space. 64.8 Since simple abelian symmetric spaces are of a trivial nature, the preceding justifies restricting to simple semisimple symmetric spaces R = G/J (G = IntOR), thus with simple semisimple rC,T1. They will also be studied globally; therefore J and its 1-component Jo must be distinguished. The fact that the adjoint action of Jo on N is bounded allows one to exclude the first possibility of Theorem 64.6, since a group generated by twin type J ( # (0))cannot fulfill this condition. From the third possibility reducibility over the reals can be canceled, for in that case, if Jo is considered as acting in N , the real multiplications in the N,belonging to the 1-component of the centralizer of Jo, thus to Jo itself, would form an unbounded subgroup of Jo. It is easily seen that a noncompact semisimple group has no faithful bounded linear representation. Therefore in the second and third cases the semisimple factor of Jo acts on N as a compact group; moreover, in the third case the central factor has the same property. So Jo itself is compact. J normalizes Jo and thus in the third case coincides with Jo;in the second case, if J‘ is the centralizer of Jo in J, then J/J’ is finite because the number of automorphism classes of semisimpleJois finite; and J’,being bounded, consists, up toJo, of the scalar multipliers +I a t most. ThereforeJ/Jois finite, and J i s again compact. This shows that the distinction made in 64.1.2 between Int R and IntOR may be dropped. Int R is closed in Aut R. (This is true for general symmetric spaces as well.) Therefore it may be supposed that: J is compact, and Jo as represented on N by j ( jE Jo) is real irreducible. Note then that J has the same property. By Theorem 64.3.1 N can be given a J-invariant real positive definite inner product (. .., ...), but now it is unique up to a constant factor, even if indefinite ones were allowed. This is evident if the representation of J in N is complex irreducible, but it extends even to the third case of Theorem 64.6. Any other J-invariant inner product could be written Y,,,,(Kn, n’) with K E End N, symmetric with respect to (..., ...), and commuting with all j ( jE J ) ; after

374

63-67.

SYMMETRIC SPACES

subtracting some scalar multiplier on N, K may even be supposed to have zero trace. K, however, belongs to the centralizer of J as stated in Theorem 64.6. Thus K is an element of adGJIN.Now

+

(Kn, n’) (n,Kn’) = 0,

since any j ( j E J ) leaves the inner product infinitesimally invariant. From the symmetry of K it follows that K = 0, which proves the assertion. Such an inner product on N is closely connected to the Killing form on G : J and N are mutually orthogonal in the sense of the Killing form J! , I of G because jiij’ EN, jiin‘ E J for j, j ’ E J, n,n’ E N ; splits into its restrictions $ J , $N to J, N. Since G is semisimple, $ is nondegenerate, as are $ and $ N . The latter is an invariant under J as represented in N a n d as such it is a multiple of the given inner product on N: y # 0. $ N ( 4 n’) = y(n, 4 , On the other hand, $ J ( j ,j ) < O for j E J, since Jo is compact and therefore all eigenvalues o f j ( j E J ) are imaginary. Now two cases are possible with the sign of y : Elliptic: Hyperbolic:

$N $N

is negative definite. is positive definite.

Again let Co be the semimorphism that defines G within G,,,. In the elliptic case $ is negative definite so c‘, being closed (see 38.5), is compact; then by 32.2.4 G is also compact and Co is some unitary semimorphism. In the hyperbolic case, with T extended to G,,, as an automorphism, is again an involutory semimorphism. The real algebra defined by C , is GI= J + iN, on which the Killing form of G,,, breaks into $J and $,N. both negative definite. Therefore in the hyperbolic case C1may be taken t o be C,,,. By momentarily writing G instead of G,,, and putting C = CoT = TCo C = C1T = TC,

Co = C,,, CI = C,,, it appears that

J = G,,,, N=

N = G,-,he Note that C # C,, since N # (0).

in the elliptic case, in the hyperbolic case,

in either case, in the elliptic case, in the hyperbolic case.

64.

SYMMETRIC SPACES

375

The requirement that ‘Cun,T1,or ‘CC,T1,be simple is equivalent to the requirement that ‘G,T’ be simple (which means C without a nontrivial proper T-invariant ideal) : In any case, C,, leaves every direct summand of G invariant. If C , is a T-invariant direct summand of G,then GI,un is a T-invariant direct summand is a direct summand of of Gun;moreover, GI is C-invariant as well, and GI,, G,. Thus a T-splitting of G induces T-splittings of Gunand G,. Therefore T I or ‘GC, T l implies simplicity of ‘G,T I . The converse is simplicity of rCun, obvious. 64.9 The foregoing completely characterizes the infinitesimal structure of a simple semisimple space R , that is, G (infinitesimal algebra of Int R ) , J, N , and Tacting on G.Its possible global structures are now analyzed. The Hyperbolic Case G may be supposed to generate a centerfree linear Lie group GI (e.g., take G instead of G and G I = Int C).Then G I is the homomorphic image of (connected) G by means of a local isomorphism x.According to 51.5.1 I , J generates in GI a maximal compact subgroup J , ; because of 51.5.10, any wrapping of G, can already be performed, as it were, within J , ; thus the X-original of J , is connected and contains the center of G. ThereforeJ, is the X-original ofJ,, whereas it should not contain any nontrivial center elements of G. Hence x is a global isomorphism. The admission of groups that are only locally isomorphic with linear Lie groups has not been an essential extension. Thus G is centerfree and may be supposed linear; J has to be compact and contain Jo, generated by J . SinceJ, is maximal compact, it turns out thatJ = Jo. Therefore in the hyperbolic case G is centerfree and J is connected. Note that by 51.5.12J is even its own normalizer in G . The Elliptic Case G may be considered as a wrapping of a compact semisimple linear Lie group, which by 32.2.4 is again compact and by 46.8 may be assumed to be linear. So the admission of groups that are only locally isomorphic with linear t i e groups again proves to be inessential. Therefore G is supposed to be a compact semisimple linear Lie group. J may exceed Jo. Its elements are T-invariants. The subset of T-invariants is a closed subgroup of G and its 1-component clearly coincides with Jo. Now J is not allowed to contain a nontrivial normal subgroup of G; since ‘G,T’ is simple, this amounts to saying that J should not contain nontrivial central elements of G. This is certainly required for Jo. Then the possible J’s are the subgroups betweenJ, and the subgroup of all T-invariants, and intersecting the center of G trivially. If G is centerfree, the set of T-invariants coincides with the normalizer of Jo

63-67.

376

SYMMETRIC SPACES

in G:first, it is evident that every T-invariant normalizes Jo. Second, if Z leaves J invariant, it leaves its orthoplement N invariant; thus ETZ-' determines the same J and N as T did; consequently, a"TE-' = T, Ta = 6, and, since G is centerfree, Ta = a. N

The Abelian Case has still to be mentioned in this context. It is clear, however, that the admission of groups only locally isomorphic to linear Lie groups does not cause an essential extension.

64.10. This analysis leads to the following: Definition If A is an abelian real linear Lie group, then A endowed with an A-invariant Riemannian metric belongs to the class of abelian symmetric spaces, Sy(A). For G E Alg Lie Lin Com SS, G # {0}, infinitesimally generating G, with different but commuting semimorphisms C, C,, (unitary), and T = CC,,, suppose the following:

64.10.1 Gc,un,com contains no proper ideal of C except (0). 64.10.2 T extends to an automorphism of G. 64.10.3 Gcsuncontains no center element # 1 of G.

Then with a group 64.10.4 G,*,,, between GC.,, and its normalizer in Gun, consisting of T-invariant elements and containing no center element # 1 of Gun, Gun/

GIun

respectively

Gc/Gc,un

endowed with a G,,-invariant, respectively, Gc-invariant, Riemannian metric are called elliptic, respectively, hyperbolic, (symmetric) spaces. These collections of spaces are indicated by Sye(Gc, Gun)

respectively,

Syh(G,-, Gun).

SY(GC,Gun) = SY~(GC, Cud u SY~(GC, Gun). Their elements are often supposed to be provided with the metric induced by the Killing form, as will be clear from the context. The reflection in the point p = G&,,, respectively, GC,,,,is also denoted by T (= CC,,). Theindication Cunwill be often omitted, since the structure of the space does

64. SYMMETRIC SPACES

377

not depend on the choice of C,,.Neither does it depend on the choice of C in its isotypic class. The attributes twin, inner, outer, and so on, are attached to these symmetric spaces according to the properties of the underlying G,. Remarks

( I ) Thanks to 64.10.1, 64.10.3, 64.10.4, Gun and G,, respectively, act faithfully on the defined spaces; 64.10.1 states that C,C,,do not coincide on any ideal # (0) of G ; 64.10.3 is equivalent to G , being centerfree. (2) The extendibility of Tneeds to be required for Gc only in the hyperbolic case and for C,, only in the elliptic case; for the latter this makes no difference; for the hyperbolic case it might a priori but in fact it does not, since if T is extendible to G , it can be made extendible to G by factoring out a central subgroup of C intersecting G, trivially. This needs to be verified only in the troubldsome cases of outer type D, with I even. Things proved can be summarized as follows: Theorem Symmetric spaces are locally isometric with products of abelian, elliptic, and hyperbolic spaces. Abelian, elliptic, and hyperbolic spaces are locally isometric with products of simple spaces of the same kind. Simple spaces are isometric with simple abelian, elliptic, or hyperbolic spaces. Elliptic spaces are compact; hyperbolic spaces are not. Up to a negative, respectively, positive, factor the differential metric at p = J of an elliptic, respectively, hyperbolic, simple space is the restriction of the Killing form. Hyperbolic spaces are globally determined by their local structure. For abelian, elliptic, and hyperbolic spaces R as before, Int R equals A,G,,, and G, respectively. In a simple elliptic or hyperbolic space the I-component Jo (=G,,",,) of the stability groupJ(=G*,,,,, respectively, G,,,,) at p acts irreducibly and faithfully in the tangent space a t p which can be identified with iGC,he,respectively, GC,he. If in this representation Jo stays irreducible over Com, Jo is semisimple; if it becomes reducible, there are two irreducible components, which are nonequivalent, and Jo is locally the direct product of a semisimple and a onedimensional linear Lie group and contains its centralizer within the group of all volume-preserving linear mappings in the tangent space.

The last-mentioned facts could have been read in the classification of simple semisimple real Lie algebras (Sections 52-53). As a matter of fact, the representation 8 becomes reducible if and only if a summand of type D,appears in the algebra of the maximal compact subgroup (which corresponds to Jo), and this happens with nonequivalent components.

378

63-67.

SYMMETRIC SPACES

64.11 The case of simple rGcom,T' with nonsimple Gcommerits a special

analysis. The Elliptic Case G is unitary and G = GI + C2, a direct sum, where G , , G ,

must be isomorphic by T. Locally G is the direct product of the locally simple G,,Gz.

Jo is the set of g . Tg, with g E G I . Now assume that G I and C 2 intersect trivially, for example, G adjoint or simply connected. Every coset g, g 2 Jo (giE Gi)intersects G I in one point g, Tg;,. This cross section G , of cosets can be used as a model of G/Jo if g, g,Jo is identified with g, Tgzl. On this model a . Tb (a,b E GI) acts as G, itself appears as an elliptic symmetric space with a metric induced by the Killing form. Note that here, as a homogeneous space, G/Jo is not reduced in the sense of 63.1.7. The Hyperbolic Case It may be supposed that G,, which is of the twin type,

arose from some complex G' by waiving. Therefore the hyperbolic space can be obtained as G'/G'"", where G' is to be taken as a real group. 64.12. Historical Note The symmetric spaces are a beautiful discovery of E. Cartan. The greater part of the results in this section and the next ones are his, though the present methods widely differ from his. Cartan's path to symmetric spaces was all but straightforward. He scrutinized a statement he found in the literature, which said that Riemannian spaces in which the curvature tensor is preserved under parallel transport have a constant curvature tensor, and proved it to be wrong. He then analyzed that class of spaces by sophisticated techniques of differential geometry [Bull. Soc. Math. France 54, 214-264 (1926); 55, 114-134 (1927) = Eurres I 2, 587-6601 and found that each such space bore a structure of homogeneous space. Actually, they are the symmetric spaces introduced in the present report; but this was an a posteriori discovery of Cartan. Later [ A m . Ecole Norm. 44,345-367 (1927) = (Eurres I 2, 867-9901 Cartan approached symmetric spaces from this point of view, though on several issues he relied on the results or presumed results of his former approach. Despite the fragmentary character and the many gaps in the general theory, all particular results are surprisingly correct. E. Cartan devoted many more papers to symmetric spaces. M. Berger [Ann. Ecofe Norm. 74, 85-177 (1957)] recently dropped the assumption of definiteness of the Riemannian metric. Though this more general case has been taken into account in the preparation of Theorem 64.6, Berger's classification has not been reported, since it is extremely involved.

65.

MINIMALAND MAXIMAL SYMMETRIC SPACES

379

Modern tale says that E. Cartan knew only compact symmetric spaces. This statement probably owes its origin to a misreading of a remark by Berger. S. Helgason wrote a monograph on symmetric spaces (Differential Geometry and Symmetric Spaces, Academic Press, New York, 1962).

65. M I N I M A L A N D MAXIMAL SYMMETRIC SPACES

By 64.10 a space in Syh(G,) is determined globally by its local structure. I n contrast, members of Sye(C,) depend globally on the choice of J . The smallest J is G,,,,,; the corresponding symmetric space is the only one with a connected stability group and it wraps all others of the same class. If Gunis centerfree, the largest J is the normalizer of G,.,, in Gun.Note that to get G faithfully represented in G/J one must bar nontrivial center elements of G from J . 65.1

Definition A space in Syh(G,), Sye(G,) is maximal (Syma) or minimal (Symi) depending on whether its stability group is minimal or maximal. The elements of Syh(G,) are both maximal and minimal.

65.2 Syh(G,);

G,,,,= J,

GC,,==N,

expN= N.

As a model of an element of Syh(G,) one can use GC,hcand even N=G,,,, because of 51.5.10 and the one-to-one character of exp on N . There the geodesics through 0 appear as straight lines (see 64.3.2). So in the case of Syh(G,) geodesic connection is unique. Splitting g E G, into nj ( n E N = exp N , j E J ) can be performed by taking the square root in N of g(Tg-l) = nj(n-'j)-' = t i 2 . For a moment call 7j(g) the action of g on N as a model of G,/J. To study 71 one must find 171 E N that satisfies gnJ=n,J;

in other words, one must split gn = nlj

(somej E J ) .

With the aforementioned procedure, n,2 = gn2 Tg-I, thus 65.2.1

( q ( g ) n ) 2= gn2 Tg-l.

380

63-61. SYMMETRIC SPACES

This suggests another use of the model N of GJJ. The new action is the transform 0 of the action 77 by means of the square mapping Q , Qn=n2

for n E N ;

and then 8, is defined by

W = Qrlk) Q-'. Therefore by 65.2.1 Q(rl(g)n) = g(Qn)(Tg-'); hence 9(g)n = gn(Tg-'). In particular,

for j

9 ( j ) n =jnj-' 6(n')n = n'nn'

E J,

for n'

E N.

The reflection in 1 again carries n into n-'. Applying 8(n'), one notes that: the reflection in n' carries n into n'n-' n'. 65.3 Sye(Gc); G,,,, = J, Jo generated by J,N = iGC,he,N = exp N. By 32.10 the one-dimensional subgroups of G,, cover G,,, whence the geodesics Yl(exp7n) Jo (n E D ) cover G,,/Jo because of 64.3 and the possibility of geodesic connection. Splitting g E G,, as g = nj, with n E N , j E Jo, is still possible, though no longer unique. The set ofg(Tg-') = njj-'n is still contained in N , and since every element of N has a square root in N the set coincides with N . One can still define a(g)n = gn(Tg-').

Writing n E N asf(Tf-') withf

E G,,,

one gets

%) n = g f V ( g f ) - ' ) ,

which is again in N. This shows that B(g) maps N onto N and makes 6 a transitive representation of G,, in N. The stability group Js of 6 at 1 consists of all g E G,, with g = Tg. Therefore it is as large as it can be as soon as G,, is centerfree. The results of 65.2-3 are summarized : Theorem A minimal model in the case of Syh(G,), respectively, Sye(C,), with centerfree G,, is furnished by N = exp N with N = GC.herrespectively, iCCghe, acted on by Gc, respectively, G,,, by means of 6 with 8(g)n = gn(Tg-I). In the hyperbolic case the model is also maximal; those symmetric spaces are determined by their local shape.

65.

38 1

MINIMAL A N D M A X I M A L SYMMETRIC SPACES

65.4. Maximal Compact Subgroups The time has come to prove a previously announced theorem (51.5. I I ) : Theorem In real linear semisimple G , all maximal compact subgroups are

conjugate. The theorem is a consequence of the following proposition: Proposition In a symmetric space R in which geodesicjoining is unique, the intersection of two balls with equal radii and different centers is contained in a ball of smaller radius.

+

Proof The uniqueness of geodesic joining in R implies that S(a,x) S(b,x) > S(a,b) unless x lies on the geodesic between a and 6 . Let m be the midpoint of a,b. The reflections, in m shows 2S(m, x)

= S(x, s, =

S(U,

X)

x) G &(a,x )

+ S(S, b, S,

+ S(a,s,

x)

X) = S(U, X )

+ S(b,x ) .

Supposing that S ( a , x ) ~ yS(b,x)3 :

J

E

-

Dj-,

+ Bl-j+l. p j - l + 2pj + . . + 2p, + 2p,,

Dj-I admits pj-l induced by an element of the kaleidoscope group of FCom,namely Sp,+.. . +P,+P,, hence inner. For j = 5 the other symmetries of D, cannot be extended without violating the invariance of the top weight of 6. Thus

t Cf. footnote in 62.7.

66. AUTOMETRISMS OF SYMMETRIC SPACES

387

JED~.

BI,l:

The mappingp, tt p, 1=4.

+ 2 p l , induced by SPl.Here, also, this covers the case

UP).

+

C I , I : J E A,-, D,. The symmetry of A,-, does not extend. &(M). J

C 1 , , , j# I # 2 j :

E

C,

+ C,-,.

Rigid graph. Trivial. CI,,,1 = 2j.

The graph of J admits the interchange of the summands, induced by '+pJ 'p2+ '+PJ+l . * * ' P J , ' ' + p l - l ' Thus

'PI+.

'

' '

'

ZZ(rs>.

+

J E AI-, D,. D I , , ,1 > 4: The symmetry of the graph of A,-, does not extend. Zz(M).t D4.1.

See D,,3 for 1 = 4.

D,,3:

J

E

D,

+ DI-,.

The graph symmetry is admitted. Z 2 ( M ) + Z z ( y ) with inner y

+ M for odd 1.

Note that for even I, M itself is inner.

+

+ +

+

D4,4: J E A, A, A, A,, the summands belonging to pl,p2,p3,p3 2p4 p, + pz. These four rootforms are orthogonal to one another. Half their sum is a rootform p, + p z + p 3 +p4 (the top weight of 8).S,,, Sp2,Sp3,

+

t This result deviates from Cartan's.

388

63-67.

SYMMETRIC SPACES

s,,3+2,,4+,, I t f 2 9 and S, I+p2+,,3+,,4 produce all rootforms from them. Therefore the full permutation group of p I , p 2 , p 3 , p 3 2p4 pI p 2 is admitted. The symmetric group of four permutands; the inner automorphisms produce the four-group.

+ +

+

The symmetries of both summands extend. y : p I t p p 2 (symmetry of DJ and ySpJt... t,,l+,,l Sp,+.. .+,,I+P2. No more for I # 2 ( j - 2), even if j = 6 or I - 2. For I = 2 ( j - 2) the aforesaid group increases by p':

I # 2 ( j - 2) :

Z,(y)

+ Z2(y + p>;

extension of the foregoing by means of p' which I = 2 ( j - 2) : interchanges the summands. Inner part: Z2(/?) Z2(p').

+

E6,1:

J E DI

+ D,.

The symmetry of D, does not extend. Z*(M). E6,2:

J E A , +A,.

The symmetry y of E,. Z,(Y).

The symmetry of D, does not extend (confront it with d). Trivial. E7,*:

J E

Dl + E.5.

If the first summand is fixed, then p4 is fixed, and so on. Z2(M).

Note that M is inner.

66. €7.3:

AUTOMETRISMS OF SYMMETRIC SPACES

389

J E A,.

The symmetry of A, extends, since the system of rootforms contains the configuration

PI

i-

ip3

.-.-

._.-._.-. PS

P7

P6

P4

P2

P2

- P3 - 2P4 - p5 - 2P6 - 2p7 P1 +2p3+p4+2p5+2p6+3p7

the symmetry of which produces an automorphism, necessarily inner in E,. Z,(P)* E8,,:

JEE,+A,.

Rigid graph. Trivial. €8,2:

J ED,.

The symmetry p4 f-) p6 does not extend. Trivial. F4,j:

J E8,

or

C, + A ,

A rigid graph in all cases. Trivial.

G,,,:

J E A , +A,.

The permutation p, tt 3p, + 2p, does not extend. Trivial. 66.8. The Autometrism Classes, Aut R/Int R, for Syma(G,), G Centerfree, Simple, t Outer Type. The Automorphism Classes of G , A,,o.*:

Jdoes not admit outer automorphisms.

zm. t Cf. footnote in 62.7.

63-67.

390 A2m-l.m.*:

J E

SYMMETRIC SPACES

Dm.

Here D, admits of a symmetry. If extendible, it must be inner up to T.

T = hP,

( k # m),

pk(h) = 0

h = exp h,

Pep = epp,

Pp, = p,+,-,,

pm(h) = Ti.

One tries

s = exp ~ ( 2 ,-t- ~z-,,),

T =

7ri/d2(pm,p,).

Then = e-,,,

Se,,

Se-,,

= epm,

Sh,,

= hpm,

SP = PS. p d s h ) = (Spm)(h)= --mi, pm-l(Sh) = (Sprn-l)(h) = @ , - I +

prn)(h) =

(i # rn - 1, m,m

pi(Sh) = 0

= pm+l(Sh),

+ 1).

However, S does not commute with T as it should. It has to be changed to satisfy this condition. Put

ho = h - Sh,

ho = exp h,.

Then pm(h0) =

27%

p,-,(ho)

STS-'

--

= pm+,(ho)= 4 N

= ShS-'P = ShP =

,

otherwise = 0;

Shh-' T = A,-'

T,

TS = T exp ~ ( 2 ,+~t-,,,) = exp

+ + 2-,,,,)) T ;

T(T(ep, e-,,))"T

= exp(-T@,,

thus STS = T,

S2= STS-I STS = Lo-'.

Now S is going to be replaced by some S'=L,S which commutes with T. This means that h, has to fulfill N

L,ho-'ThT1 = 1

or, with h l = exp h,, h, - Th, = Lo mod 27ri

for the rootforms.

67.

FUNDAMENTAL GROUPS OF SYMMETRIC SPACES

39 1

From this hl can be solved with p,&)

=

~ , , , - ~ (= h -irk ~)

0,

~ ~ + ~ =+Ti, ( h , )

otherwise 0.

Now S' commuting with T leaves Jinvariant and produces the automorphism wanted. Thus Z2(T) + z2m D l P j , * ,I# 2j - 3. No automorphism of the graph of J . ZZ(T)* J

D,.j,*, I = 2 j - 3 :

E

Bj-2

+ Bj-2.

The graph admits the interchange of the summands. Put s=sp,+...+pj

... ~P,-Ii-...+pl-ls p j + . . . + p l *

The rootforms are permuted in the correct way by S. Moreover, SP = PS, but Sdoes notcommute withT= t?P.Thecorrectingfactorisfoundas inA2,n-lvm,*: PpI

= p2,

Pp2 = pI, Pj(h) = Ti,

pj(Sh) =-Ti,

ppk

= pk

Pk(h) = 0

(k # 1,2),

Pep = epp,

( k #h

pl(Sh) = p2(Sh) = r i ,

otherwise 0,

and so on: TS = S - I T,

and so on. Z2(T) + ZAP). E6.0,*

and

'6,?.,**

No graph automorphism. Z*(T).

67. FUNDAMENTAL GROUPS OF SYMMETRIC SPACES 67.1 Hyperbolic symmetric spaces are topologically euclidean, with a trivial fundamental group. For elliptic symmetric spaces of the twin type the fundamental group is that of the unitary type. The case R E Syema(G,) is settled by Theorem 6 3 . 3 . 4 . Indeed, R = GUn/Cc,,,, where Gc,,, is the maximal

392

63-67. SYMMETRIC SPACES

connected subgroup of Gunconsisting of T-invariants. I f j , as in 63.3.4, means the embedding of G,,,, into Gun,then the a E~@(G,,,.) are characterized by containing a T-invariant path. Now for centerfree Gunthe system of straight paths defined by 2 in 32.2.4 intersects every element a of @(Gc,un)at most once, and this system is T-invariant. Therefore for centerfree G,,, hence, in general, every CL = Ta contains a path w = Tw. Whence: Proposition For R = Gun/Gc,un, @(R) is isomorphic to @(Gun) mod kernel (1 - T ) . Its dual @*(I?) is isomorphic to (1 - T ) @*(Gun). 67.2. Dual of the Fundamental Group of R E Syema(G,), G Simple. Centerf ree

For all inner types: Trivial.

Dlv*:

Elements rl + r2and xk (k > 2).

I odd : Z 2 ( 2 ~ ,= ) Z2(~3), 1 even: Z2(r1+ r2)= Z2(~3).

67.3. Proposition Let G be simple centerfree, R E SyemXG,), Fo = Gun, Jo = Gc,,,,, and J the normalizer of J o in Fo. Then (P(R)/@(Fo/Jo)is isomorphic with J/Jo. This is evident since Fo/Jo is a wrapping of R = Fo/J. J/Jo is found in 66.7-8; the normalizer classes produce the isometrisms indicated by p, p’, and M (if it is inner). If J/Jo or @(Fo/Jo)is trivial, the proposition gives full information on @(R). Otherwise some additional arguments are needed. 67.4. Fundamental Group of R

AlSj, I # 2 j - I : 1=2j- 1 : Bl,j, j > 3 : B1.1: BL2: c1.1:

E

Syemi(G,), G Simple, Centerfree

67.

393

FUNDAMENTAL GROUPS OF SYMMETRIC SPACES

Proof for A2,,-, ,,.* : The notations of 66.8 are used. The fundamental group can be considered as an extension of Zm(n2)by Z2(P). A path from 1 to I?, S i n the normalizer ofJ, is found as the adjoint image of a path

Y,

exp th, exp tT(e,,

+ e-,,)

(0 G t G 1)

in the universal wrapping of Fo = Int R. The square of the path is

+ e-,,)

(0 G t G 2).

Y, exp +tho exp tT(e,," + e-,,)

(0 G t G 2)

Y, exp rh, exp tT(e,, The path

has its adjoint within J,. Therefore in Fo/Jo it is trivial. Multiplication of the last but one by the inverse of the last gives

Y, exp t ( h , - $ho)

(0 G t G 2).

Now pm(2h,- h,) = -27ri, pm-,(2h1- ho) = 0, and - 0 otherwise, which shows

x 2 is a

pm+,(2hl- h,)

= 2ni,

generator of order m of the dual fundamental group of Fo/J,. Thus,

Y,

exp r(h, - tho)

.I This restilt deviates from Cartan's.

(0 G t G 2)

394

63-67. SYMMETRIC SPACES

turns out to be an element of order m;

+

(0G t G 1) y, exp th, exp t+,, e-,,,,) is of order 2m in the fundamental group of Fo/J. For Dl,,,* ( I = 2 j - 3) the proof is much the same. 67.5 In the inner case all wrappings of R E Syemi(G,) (G simple, centerfree) are realized by elements of Sye(G,). An element of Syema(G,) is simply connected. So wrapping Fo = G,, does not affect R, since the arising center dives into the stability group. In the outer types R E Syema(G,) is not simply connected (67.2). Therefore wrappings of G may cause true wrappings of R. Let be the universal wrapping of G and = Po, = Jo the induced wrappings of G,,, G,, G,,,,, respectively, within The new symmetric space k = Po/Jo will wrap R in any case; however, Jo may have a nontrivial intersection with the center of Gun. This intersection 2 can be found by comparing the results of 62.8 and 62.9. The 1-component of Aut k is isomorphic to f o / Z .

e,,

Proposition

e

e,, e,,,,,, e.

4 is simply connected.

Proof It suffices to show that the index of Z in the center of go equals the order of the fundamental group of R. This is shown in the list in 67.6, obtained from 46.4, 67.2,62.8, and 62.9.

67.6 A,: center of Po: z,+*(TI). @(R): Ztcl+,,(27r1)for odd Z, Z,+,(7r1)for even 1. 2: Z2(7ri) for Al,o,*,odd Z, 0 for even 1. Z2(7rJ for A2m-l.m.*,odd m yZ2(7r3,even m. D,: center of Po: Z2(7rl) Z2(7r2) for even Z, Z4(7r1)for odd 1. (@R): z2(r3). 2: Z2(74*

+

E, : center of go: Z3(7rJ @( R) : Z3(77,)* Z: 0. Remark The validity of the proposition is also an easy consequence of 64.1.3

and the considerations in 64.9. 67.7 Aut R/Int R can sometimes be smaller than that of the maximal space. One must check the action of Aut R/Int R on the fundamental group of R E Syema, in other words, on J/Jo as far as the inner types are concerned. In the outer case there is still the a priori condition that T be an autometrism. It turns out that only for D2(,-2),,and the outer types of D, can it happen that Aut R/Int R is smaller than for the maximal type.

TITS GEOMETRIES

68. A LIST OF FUNDAMENTAL THEOREMS

The aim of this section is to list a series of closely related theorems, a few known to the reader, one to be proved here, and the greater part in the next section. 68.1-14

G E Alg Lie Com, and if needed generated by G.

E

Alg Lie Lin Corn; then G is

Notation IfX sub G,thenInt(G, X)isthesubgroupofinnerautomorphisms of G leaving X invariant and f is the group generated by the exp X with x E X as acting on G. 68.1. Trunks

A regular element of G was defined as one in which rank 2 (x E G ) attains its maximum. The trunk H of a regular element h was the set of x belonging to the 0-root of h, that is, hPx= 0 for large p . The following is well known from 17.8 and 33.9: 68.1.l.Theorem All trunks of G are Int G-equivalent. 68.1.2 If Cis semisimple and His some trunk of G, then His the subgroup of

Int G leaving H elementwise invariant and Int(G,H)/A is isomorphic with In t W *(G,H ) .

68.2-3. Maximal Solvable Subalgebras 68.2 The maximal solvable subalgebras of G are of particular importance. They are also called Borel algebras (after A. Borel). Theorems 68.2.1 A maximal solvable B sub G contains a trunk of G. 68.2.2 The maximal solvable subalgebras of G are Int G-equivalent. 395

396

68-75.

TITS GEOMETRIES

68.2.3 If B is maximal solvable sub G, then Int(G, B ) = B. 68.2.4 For semisimple G an element of Int(C, B ) leaving B elementwise invariant is the identity. 68.2.5 The groups infinitesimally generated by maximal solvable subalgebras of G are closed and characterized by being maximal solvable Lie or maximal solvable connected. They are Tnt G-equivalent and their own normalizers. 68.3 If G E Alg Lie Com S S and H i s an ordered trunk of C, then clearly H , together with the branches e, ( a > 0), spans a maximal solvable subalgebra. The next theorem asserts that this construction exhausts all of them. Theorem If G E Alg Lie Corn S S , then any maximal solvable subalgebra

B of G is spanned by an arbitrary trunk H of G within B and the branches e, with a > 0 under a suitable order on H:. The maximal solvable subalgebras of Garound a trunk H a r e Int(G, H)-equivalent and in a one-to-one relation with the chambers on H .

68.4-6. Nonsemisimple Maximal Proper Subalgebras 68.4. Theorem G E Alg Lie Corn S S . Any maximal proper subalgebra of G, which is not semisimple, contains a trunk of G. The case of Bl-l in D, shows that a semisimple G can have semisimple maximal proper subalgebras with no regular element. 68.5. Definition Let G E Alg Lie Corn S S be dressed on an ordered trunk H . For p E W++the set of CL E W + with a vanishing p-coordinate on the basis W + +is denoted by W'(p). The linear space spanned by H, all e, with cc > 0, and all e-, with cc E W+(p) is denoted by G ( p , H ) or C(p),for short. 68.6. Theorem G E Alg Lie Com S S . 68.6.1 C(p),as defined in 68.5, is a maximal proper subalgebra of Cand nonsemisimple. 68.6.2 Any nonsemisimple maximal proper subalgebra M of G on any suitably ordered trunk H of G within M gets the form G ( p , H ) for some p E W++(G,H).

68.

397

A LIST OF FUNDAMENTAL THEOREMS

68.7. Parabolic Subalgebras 68.7.1. Definition The proper subalgebras of G containing a maximal

solvable one and the groups generated by them are called parabolic. Note that a parabolic subalgebra of G must contain rad G. 68.7.2. Theorem If M is parabolic sub G, then Int ( G , M )=

a.

Parabolic subgroups are closed and their own normalizers. 68.7.3. Theorem Parabolic M sub C i s the intersection of (a finite number of) maximal parabolic subalgebras of C around the same trunk, and likewise for parabolic subgroups of G. 68.7.4. Theorem For semisimple G the notions of maximal parabolic and of nonsemisimple maximal proper subalgebra coincide. 68.7.5. Theorem For any ordered trunk H every maximal parabolic subalgebra of semisimple G is Int G-equivalent with some C(p). For any ordered trunk H , C(p,H ) and G(o,H ) are Int G-equivalent only if p

= u.

68.8. Nilpotents, and Semisimple Subalgebras of Rank 1 68.8.1. Theorem To any ad-nilpotent e E G E Alg Lie Com SS an ad-pure h E G can be found such that he = e. 68.8.3. Theorem Any ad-nilpotent e E G E Alg Lie Com S S is contained in a semisimple subalgebra of G of rank 1. 68.9. Bruhat’s Lemma The intersection of two maximal solvable subalgebras of C E Alg Lie Com S S contains a trunk of G. 68.10-1 1. Some Homogeneous Spaces 68.10 Int Gacts transitively on the set of maximal solvable subalgebras of G. There is even double transitivity in a weak sense: Theorem

Let G t Alg Lie Corn S S . Then the (ordered) pairs of maximal solvable subalgebras of C which have no more than just a trunk in common are Int G-equivalent and form an open everywhere dense subset in the manifold of all (ordered) pairs of maximal solvable subalgebras of C .

398

68-75.

TITS GEOMETRIES

68.11. Theorem Let G E Gru Lie Lin Com, A a maximal solvable Lie sub-group. Then the manifold G/A is compact. 68.12-14.

A Proof Only the theorems of 68.2 are proved this section. The proof 68.12 rests on the rather elementary fact (to be proved in 69.25 and 69.37) that: the maximal solvable subalgebras of G with a regular element are equivalent under Int G and form a compact manifold, and on a proposition that for the sake of convenience is formulated in projective terms. A group acting on the linear space R can also be interpreted as acting on the projective space R , derived from R. Lie’s theorem (13.9) then means that a solvable Lie group possesses a fixed point. Proposition Suppose that G E Gru Lie Lin Com, solvable, acting on R E Spa Lin Com, interpreted as acting on R,. Then any minimal closed Ginvariant subset of R , consists of one point.

The truth for dim R , = 1 is a matter of inspecting the solvable linear Lie groups acting on two-dimensional space. From a triangular form of G it is seen that either every point of R , is a fixed point or that only one point of R , is left fixed and all others form one orbit, from which the validity of the proposition is obvious. Therefore suppose that dim R , > 1 and let M c R , be such a minimal closed invariant subset. Its projective span is again invariant and may therefore be supposed to coincide with R,. As acting on R,, solvable G possesses a fixed point p. Ifp E M , then because of its minimality Mcoincides with {p}, which agrees with the assertion. If p $ M ,then in Rk = R , m o d p (i.e., the projective space of straight lines through p) M reappears as a closed subset M’. It is again minimal invariant under the solvable group induced by Gin R>. By induction M’may be supposed to consist of one point only. This means that M lies on a straight line through p, and finally that dim R, = 1, which is just the induction basis. Proof by induction on dim R,.

68.13. Proof of 68.2.1. Let9lbethesetofmaximalsolvablesubalgebrasof G with a regular element and let A be any solvable subalgebra of G. The ii (a E A ) generate a solvable linear Lie group A’acting on the linear space Gand consequently on 93. Now 93 and the action of A’ on 9l can be interpreted in a linear space R (and in the related projective space R,) as follows: Let dim B = k for B E 9land take R as the linear space of skew k-tensors on G. (Using a basis of G, the k-dimensional linear subspaces L of G are described by their Plucker coordinates in R ; take a basis of L and form the k-determinants from the coordinates of the basis elements.) Now A’ induces a solvable

68.

A LIST OF FUNDAMENTAL THEOREMS

399

linear Lie group A" acting on R and consequently on R,. The image of93 in R , is again closed and A"-invariant. By 68.12 it contains some element of 93 fixed under A". This means the existence of a solvable subalgebra B with a regular element such that a"B= B for a" E A', hence a'B c B for a E A . Now the linear span of B and any a E A is again solvable; therefore a E B for any a E A , hence A = B. If, moreover, A is maximal, then A = B which proves that every maximal solvable subalgebra has a regular element. The remainder of the theorem is an easy consequence verified in a later context. 68.14 The facts proved in 68.12-13 are not used in the discussion of the theorems stated in 68.1-11. The fact (68.2.1) that a maximal solvable subalgebra of Gcontains a regular element is related to Lie algebras rather than to groups. For this reason it will be proved algebraically in Section 69, together with the analogous assertion on maximal nonsemisimple subalgebras of semisimple Lie algebras (68.4). From this the conjugacy assertions will be derived by using the conjugacy of trunks. 68.15. Historical Note The statements on the existence of regular elements in maximal solvable and nonsemisimple maximal subalgebras and some others, such as 68.8.1 and 68.8.3, are being credited to V. V. Morozov on account of his unpublished Kazan thesis of 1943, which seems to have been lost. Without doubt it is a great merit to have proposed these theorems and to have initiated this new chapter of the theory of Lie algebras. On the other hand, it is a fact that Morozov's published work on these points contains nothing that by any standard whatsoever could be called a mathematical proof [Dokl. Akad. Nauk. SSSR 36,83-86 (1942)l. F. I. KarpeleviE [Dokl. Akad. Nauk SSSR 76, 775-778 (1951)l is sometimes quoted as having simplified Morozov's proofs, though rather KarpeleviE used Morozov's full results to derive statements that in Morozov's strategy must have appeared as lemmas from which his final results were derived. It is distressing that until now no serious proof for a series of primordial theorems, though often quoted and used, has been available in the literature. Exceptions are Morozov's rather weak statements 68.8.1 and 68.8.3, which were first proved by N. Jacobson [Proc. Amer. Math. Soc. 2, 105-133 (1951)l. The proof of 68.2 in 68.12-13 has been fashioned after one for algebraic groups, thanks to an oral indication by J. Tits; 68.10-1 1 likewise go back to J. Tits [Mem. Acad. Bruxelles in 8", 29, No. 3,94(1955)]. Bruhat's lemma (68.9) seems to have been diffused by oral tradition. Its usual formulation and proofs have concealed the fact that it is almost trivial, at least for linear Lie groups and algebras. For an axiomatic and more detailed description of the situation as in 68.7 see J. Tits [Compt. Rend., Paris 254,2910-2912 (1963)l.

68-75. TITS GEOMETRIES

400

The theorems announced in the present section are proved in the next, together with other statements. The proofs, except those mentioned before, are due to H. Freudenthal, though simplified by H. de Vries. 69. PROOFS OF THE STATEMENTS OF SECTION 68

G E Alg Lie Corn.

tr(ab) is a linear 69.1. Proposition If G E Alg Lie Lin Com S S , then combination of the Killing forms of the simple summands of C, with positive coefficients. Proof According to 43.4.1, this is true of simple C, even if reducible. Assume G = GI G2 (direct sum) and the proposition true for every nonzero linear

+

representation of GI and G,. It is sufficient to prove the proposition for irreducible G. From 45.3it is easily seen that up to equivalence any irreducible representation of G, as well as the identical one, originates as follows: let Gi act on Ri E Spa Lin Com, i = 1, 2, and let G act on R = R 1 Q R2 in such a way that for ai E Gi, xi E Ri (a1

c1 E End R ,

+ a2)(x1 Q ~

2= ) (aixi)

Q ~2

+XI Q (a2~2).

induces c E End R, where tr c = (dim R,) tr cl.

a l , b , E GI induce a,b E G acting on R,

trR Ub = (dim R2) trR,

bl.

a1 E GI, b2 E G2 induce a,b E Gacting on R,

trRab = tr a I * tr b, = 0. Hence

+

+

tr(a, a2)(b1 6,) = dim R, tr a1b1 + dim R,tr a2b,. This proves the assertion. 69.2. Generalization of the Second Criterion on Solvability (17.17) Proposition For G E A I ~Lie Com Lin the following conditions are equivalent: (i) G is solvable, (ii) tr ab = 0 for a E G, b E C ( G ) ,(iii) tr ab = 0 for a,b E C(C). Proof (i) --f (ii) and (ii) + (iii) are clear. Now suppose that G is not solvable. G acts on R which can be written directly as R I * R, such that for every i

+- -+

69.

R, + * R,+ *

*

PROOFS OF THE STATEMENTS OF SECTION 68

401

- + R, is G-invariant and + R,,, mod R , + - + Ri is irreducibly acted on by G.

* *

Since G is not solvable, dim R,> 1 at least once. Let t , be the projection onto R,, thus t , x = x . If a’ E End R for a E C i s defined by u’x =flax for x E Ri and all i, then You’ maps G homomorphically onto some C’ and C ( G ) onto C(C’). Since G’ is irreducible on every Ri and dim R, > 1 at least once, it appears that C(G’) is simisimple and # (0). Thus tr a‘b‘ does not vanish identically on C(C’)according to 69.1. The same is true of tr ab, which equals tr a’b’.This proves the assertion.

xi

69.3. Proposition If G E Alg Lie Corn Lin and N is the subalgebra of nilpotent elements of rad C, then N is an ideal of G, and rad(Cmod N ) = (rad G ) mod N is the center of G mod N .

From a triangular form of rad G it is clear that N is an ideal of rad C, containing the commutator algebra of rad C . Let g E G. Then g and rad G span a solvable subalgebra whose commutator algebra consists of nilpotent elements and is contained in rad G. Therefore 2 rad G c N , for all g E C, from which the proposition follows. Proof

69.4. Proposition If G E Alg Lie Com Lin and R is the subspace consisting of all elements r such that tr ar = 0 for all a E C , then (i) R = rad C, (ii) R contains the nilpotent elements within rad G, (iii) R is an ideal of C, (iv) (rad C ) mod R is the center of C mod R .

R is an ideal since tr(a[b, r]) = tr([a,b]r) = 0 for a,b E C, r E R . The solvability of R follows from 69.2; hence R c rad C. If a E G, than a and rad G span a solvable subalgebra, from a triangular form of which it follows that tr an = 0 if n is a nilpotent element of rad G. Hence R contains the ideal N of nilpotent elements within rad C. From 69.3 it follows that (rad G ) mod R is central in G mod R ; therefore (rad G ) mod R = rad(G mod R ) is the center of Cmod R. Proof

69.5. Proposition For A sub G, the following conditions are equivalent: (i) A is solvable, (ii) #(a, 6) = 0 for a E A, b E C ( A ) ,(iii) #(a, b) = 0 for a, b E C(A).

Apply 69.2 to the image of A under YaeA YgcG i g , that is, to ad,A; note that the kernel of this representation is A n Z, where Z i s the center of G and A n Z c rad A. Proof

402

68-75.

TITS GEOMETRIES

69.6. Notation If M sub G and z,h is the Killing form of G, then

M I = the orthoplement of M with respect toL,I and MO=ML n M Proposition If M sub G, then (i) M o c rad M , (ii) M u contains the adnilpotent elements within rad M , (iii) M n is an ideal of M , and (iv) (rad M ) mod M u is the center of M mod M u . Proof Apply 69.4 to the image of Munder the representation

YmEM YgeC

&g;

note that its kernel lies within Mu. 69.7. Proposition I f M sub G E Alg Lie Com and adGMadmitscleaving, then a d G M oalso admits cleaving. Proof Since M n is an ideal of M , f M c M u for x E M u . By the last assertion

of 18.1.I the cleaving components o f f also map Minto M u ; let them be pure ii and nilpotent E (some u,e E M ) . This shows that e M u is in the center of M mod M u ; hence by 69.6 (iv) in (rad M ) mod M u , thus e E Mu by 69.6 (ii), which proves the assertion.

+

69.11 . Proposition Suppose G E Alg Lie Com Lin, G solvable, and let G

admit cleaving. Let N be the ideal of nilpotent elements of G. Then, if A is a maximal pure subspace of G, G = A N, a direct sum, of subspaces.

+

+ N. Therefore, if G # A + N, then by the fact that the elements of A are ad-pure there would be a g E G, g 4 A + N, and u E A such that iig is a nonzero multiple of g;this contradicts the fact that Ncontains the commutator algebra of G. Proof By 19.20 the idealizer of A is contained in A

69.14. Proposition If G E Alg Lie Com S S , H a trunk of G, and A c H, then the centralizer of A is the direct sum of a subalgebra of H and a semisimple one of G. This follows from 20.15. 69.15. Proposition Let G E Alg Lie Com S S , M sub G its own idealizer, and A a maximal ad-pure linear subspace of M . Then z,h is nondegenerate on A . Proof According to 19.22, A lies in a trunk H of G. For a

E W the restriction to A of a is denoted by a’.The class of 6 E Wwith 6’ = a‘ is denoted by [a]. G splits with respect to /I E H into eigenspaces G, (a E W).

69. PROOFS OF THE STATEMENTS OF SECTION 68

403

G splits with respect to a E A into eigenspaces G,, ( a E W ) .

Put

M n C,. = Ma,.

M=

at

Ma,,

all sums being direct. Let B be the set of h E Hsuch that a(h) = p(h) for all a,p with a' A c B. It will be shown that A = B. Every x E C,, has the form x= x, with x, E C,.

=

p'. Thus

YE[=]

ForhEB hx =

c

y(h)X,

yaal

= a(h)

c

X, = a(h)X.

Thus

BM,, c Ma,,

BM c M ,

and since M is its own idealizer, B c M . Hence B c M n H = A , and thus A = B. Now A can be singled out of H a s the set of h fulfilling a(h)= P(h) for certain pairs a,p E W . Thus A is spanned by rational linear combinations of nodes and # is nondegenerate on A by 21.2. 69.16. Proposition If G E Alg Lie Com SS and M sub C is its own idealizer, then Mu consists of the ad-nilpotents within rad M. Remark This is a stronger version of 69.6 (ii) in this special situation. Proof Since M is its own idealizer, it follows from 18.1.1 that the cleaving components of elements of adcM are again in ad,M; therefore cleaving is possible in ad&. By 69.7 cleaving is possible in adcMn. Hence by 69.1 1 M u = A l + N is a direct sum of subspaces, in which A , is ad-pure and N consists of the ad-nilpotents in MU. Let A be a maximal ad-pure subspace contained in M and containing A ,. By 69.15 I) is nondegenerate on A . But #(A, A , ) C #(M, M u ) = {0},whence A l = (0).This proves the assertion.

69.17. Proposition Suppose that Nsub F sub G E Alg Lie Com, N is adnilpotent in C, F is not ad-nilpotent in G, and ad-cleaving with respect to G is possible in F.Then some ad-pure x f O from F fulfills #(x, N ) = (0).

404

68-75.

TITS GEOMETRIES

Proof N may be supposed to be maximal ad-nilpotent contained in F. When

m

looking for an eigenvector of the representation induced by in F mod N , one F\ with mx c N. Because of the maximality of N , such an x is finds an x E N not ad-nilpotent and after cleaving it may even be taken ad-pure. It clearly fulfills $(x, N ) = (0). 69.18. Proposition Suppose that G E Alg Lie Corn S S , M sub G is its own idealizer, M’ c M , and A is a maximal ad-pure subset of M . Then A is a trunk of G. Proof The notation of the proof of 69.15 is used, in particular a trunk Hof

G with A c H c Go’.Then by 19.20 applied to :,dcM 69.18.1

Mo, = A

+ N,

a direct sum, with ad-nilpotent N.

By 69.14 69.18.2

Go,= A ,

+ G’,

a direct sum, with A c A , c Hand semisimple G:

One must now show G’ = (0).

A , =A,

(Then indeed A = H.) Suppose G‘ # (0). Since (69.18.1), N c Mo, c Go,,and N is ad-nilpotent, N c G’ (see 69.18.2). By 69.17 there is an ad-pure x # 0, x E G‘, such that $<x, N ) = (0). x may even be supposed to be in H, although H may have to be modified, with-

out impairing A , . $ ( A , N ) = (0) because of 69.18.1. Thus $(x

+A,N)

=

(0).

Since $ is nondegenerate on A (see 69.19, x may be replaced mod A by x’ such that moreover $(x’,

4 = (0).

Since x’ E x + A ,

$W, N ) = (01. Thus

$b’,MO!)= (01 for some x’ E Go,\Mo,

thus for some x’

I$

M.

69.

PROOFS OF THE STATEMENTS OF SECTION

68

405

Now #(G,,,,Ga,)= (0) for a’ # 0’, and Ma, c Gm,;thus #(x’, M a , )= (0)

for

OL’

# Of,

#(x‘, M ) = (01,

but x’ E M I contradicts x’ # M because of the supposition that M I c M . This proves G’ = (0). Now suppose that A is a proper subset of A , . Then, since # is nondegenerate on A (see 69.19,

for some x‘

#(x‘, A ) = (0)

E A,\A.

By the same reasoning this leads to a contradiction, which proves the statement. 69.19. Proposition If G E Alg Lie Com S S and M is maximal solvable, then M I c M . Proof Suppose that it is not so. Then the f (x E M )have a simultaneous eigenvector in M I mod Mu. Thus for some c E M’-\Mand some X

(2- X(x)) c

E

for all x E M .

Mu

+(c, (a - X(X)) c )

E

+ ( M I ,M U ) = (0).

#(c, ac) = -#(c, Ex) = #(Ec, x)

= 0.

Thus

h(x)+(c, c) = 0. A = 0 would mean .?c E M for all x maximality of M . Therefore

E

M , thus EM c M, which contradicts the

#(c, c) = 0.

Let Kbe the algebra spanned by M a n d c, L that spanned by M u and c. Then L is ideal in K, and K mod L is solvable. So L must not be solvable. However, # ( M n , M u ) = #(Mu, c ) = (0)

and

#(c, c ) = 0,

which makes L solvable. The assertion is proved. 69.20. Proposition If C E Alg Lie Com S S , M a maximal proper subalgebra of C,and Mnonsemisimple, then M I ,= M . Proof M is its own idealizer and M I is invariant under N

particular, under M u .

A,therefore,

in

406

68-75.

TlTS GEOMETRIES

Suppose that M u # (0). By 69.16 M u is ad-nilpotent. If the proposition were not true, then for somep zz 0

M,

$PMI+

f i ” + ’ M l C M.

On the other hand, c n;lMl c M I

ji&/fl

and by induction N

Mnp+l M

I cMI.

Thus N

M O p + l MJcMU, N

N

M q M o p M * ) c MU, N

hence M o P M Lin the idealizer of M u . Now the idealizer of M u is nonsemisimple and proper sub Csince M u was ad-nilpotent and supposed to be # (0). Because of its maximality, Mcoincides with the idealizer of MU and therefore N

M O P M l c M , which contradicts

the definition ofp. To complete the proof that M I c M , one must exclude

69.20.1

MU

= (0).

Here the semisimplicity of G must be used. If 69.20.1 were true, then by 69.6 (ii, iv) rad M would be the center of M a n d no ad-nilpotent # 0 would be in rad M . Since M is maximal, it is its own idealizer; hence ad-cleaving subsists in M a n d in its center (see 18.1.1), that is, in rad M . Therefore rad M is ad-pure. Under rad M , in the adjoint action, G splits into eigenspaces G, with M = Go.Since rad M # {0}, one can choose CL # 0 such that G, exists and G2, does not. Then Go G, is larger sub G than M, but it does not yet exhaust G, since rad(Go G,) = (rad M ) + G, so that Go+ G, is not semisimple. This contradicts the maximality of M and disproves 69.20.1.

+

+

69.21 . l . Proposition A maximal solvable subalgebra ofC E Alg Lie Corn contains a trunk of G. 69.21 -2. Proposition A maximal proper subalgebra of G E Alg Lie Com

SS which is not semisimple contains a trunk of G. 69.21.3. Proofs Proposition 69.21.2 follows from 69.18 and 69.20. Likewise 69.21.1 follows from 69.18-19 with respect to G’ = G mod rad G and from 17.12.

69.

PROOFS OF THE STATEMENTS OF SECTION

68

407

69.21.4. Remark Propositions 69.21.l-2 contain Theorems 68.2.1 and 68.4. 69.23. Definition If G E Alg Lie Com SS, H a trunk of G, M s u b G spanned by a subset of H and a set of branches belonging to H , then a E W* is calledfvee (with respect to M ) if

e,EM

and

e-,$M.

Proposition Under a suitable order on H,T all free rootforms are positive. Proof Instead it will be shown that a suitable element of the kaleidoscope

group maps all free rootforms into positive rootforms. For a proof by induction on the number of positive free rootforms and along the order in H: it is sufficient to show that by means of the kaleidoscope group every maximal negative free rootform can be raised without changing the signs of the positive free rootforms. First remember that for p E Wti. all rootforms except k p preserve their signs under S,. Let -a be a maximal negative free rootform. There is a p E W + + such that Sp(-a) > -a. If p is not free, then Sp also maps the positive free rootforms into positive rootforms. So p may be assumed to be free; but then, withe-,,e, E Malsoe,-, E M(note that ofnecessity a # p). This contradicts the maximality of -a, since if p - a were not free, then with ea-p and e, also e, would be in M , in contradiction to the fact that -a is free. 69.24.1. Proposition Any maximal solvable subalgebra M of G E Alg Lie Com SS is spanned by an arbitrary trunk H c M a n d the branches e,, with a > 0 under a suitable order of H;. Proof By 69.21.1 Mcontains a trunk, say H . By 20.15 it fulfills the conditions of 69.23. Since e,, e-, generate a semisimple subalgebra, it is not allowed that

both of them belong to M . Therefore, if e, E M , then a is free. By 69.23 an order exists such that a > 0 for each e, E M . The subalgebra spanned by H and all e, with a > 0 is still solvable. Therefore it coincides with M, which proves the assertion. 69.24.2. Proposition Among the algebras M s u b G E Alg Lie Com SS around H the maximal solvable ones are characterized by the following property : for every a E W* exactly one of the branches e,,e-, belongs to M . This is a consequence of 69.24. I or proved in the same way. 69.25 The maximal solvable subalgebras around H are thus one-to-one related to the chambers in H:. By 33.4.3 and 33.9(4) they are equivalent to one another under Int(G, H ) . Further, since two trunks are equivalent under

408

68-75.

TITS GEOMETRIES

Int G, one gets the Int G-equivalence of maximal solvable subalgebras and, as in 69.21.3, the following more general proposition. Proposition All maximal solvable subalgebras of G E Alg Lie Com are equivalent under Int G. Remark This is Theorem 68.2.2.

69.26 Again suppose first that G is semisimple. An element a" of Int G leaving maximal solvable M invariant changes a trunk H within M into a trunk. The same is achieved by some 6 E A.Then 6-'a" E Int G leaves Mand H invariant, thus the chamber in H: corresponding to M , thus every element of H. Hence &-Id E R, a" E A?. If, moreover, d leaves M elementwise invariant, then a" = exp h for some h E H, and finally it is the identity. If the method in 69.21.3 is used again,? the following is obtained: Proposition If C E Alg Lie Com, then any element of Int G leaving the maximal solvable M invariant belongs to I@. Moreover, for semisimple G, if it leaves Melementwise invariant, it is the identity on G. Theorems 6 8 . 2 . 4 are completely proved by 69.21.1-2 and 69.24-26; 68.2.5 is an immediate consequence (see 13.3).

69.27. Proposition If G E Alg Lie Com SS, then G ( p ) , as defined in 68.5, is a maximal proper subalgebra and nonsemisimple. Proof The orthoplement of C(p)is spanned by the e, such that p appears in a with a positive coefficient. It is an ad-nilpotent ideal of C ( p ) ; therefore G(p) is not semisimple. To show that it is a maximal proper subalgebra, consider the algebra G' generated by G(p)and one e-,, where p appears in a with a positive coefficient,and prove that G' = G. For GI = p this is obvious. For CL > p there is a /3 E W +such that a - BE W k and a - @ > p . Then Ege-,, whence also eg-,, in G', and by an obvious induction G' = C . (One notes that by 20.15 any subalgebra containing C(p)is spanned by H and branches.) 69.28.1. Proposition If G E Alg Lie Corn SS, M is nonsemisimple maximal proper sub G, His a trunk of G contained in M, and the order on H: is such that every free a (e, E M) is positive, then M i s some G ( p ) and thus maximal parabolic. In particular, every maximal parabolic subalgebra of G around H is Int (G,H)-equivalent with some C(p).

t Observe that the kernel of the homomorphism from Int G to Int (G mod rad G) may be disconnected. See 69.29.3.

69. PROOFS OF THE STATEMENTSOF SECTION 68

409

Proof According to 20.15, M is spanned by H and a set of branches. If a E W* and e-u $ M , then ea E M I c M by 69.20; therefore the set V of a E W* with ea E M has the property a E V or -a E V for a E W*. By the fact

that all free rootforms are positive one even has W +c V. Because M f G, there is a p E W + +such that e-,, 4 M , that is, p is free. Suppose there were a nonfree positive a 2 p ; assume a minimal. As remarked in the preceding proof, there is a /?> 0 such that a - p E W* and a - 2 p. Then EPe-,, hence also eP-., in M , so a - p is still nonfree with a - /3 > p ; this contradicts the minimality of a. It follows that M c C(p);hence M = C(p). 69.28.2 If in the foregoing M is just a parabolic subalgebra, then by Proposition 69.24 one can still assume that W +c V, where V is as above. The rest of the proof shows that M is the intersection of the C(p) with p free and primitive. Hence: Proposition Every parabolic subalgebra is an intersection of maximal parabolic subalgebras around one and the same trunk. This remains true for arbitrary G E Alg Lie Com by the principle applied in 69.21.3.f

69.28.3. Proposition Among the algebras M proper sub C E Alg Lie Com SS around n trunk H the parabolic ones are characterized by the following property : for every a E W*, eu E M or e-a E M. This is a consequence of 69.24.2. 69.29.1. Proposition If G E Alg Lie Com SS and His an ordered trunk of G, then any nonsemisimple maximal proper subalgebra M of G is Int Gequivalent to some C(p). Every element of Int G that leaves M invariant belongs to A?; an element of Int G that leaves M elementwise invariant is the identity. No C(p), G(a) with p # a are Int G-equivalent. Proof By 69.21.2, M contains a trunk H' of G. Under the action of Int G one

may assume that H ' = H . By 69.23 under a suitable order on H: all free rootforms (with respect to M) are positive. By some s' E Int G this order on HZ is transformed into the given one on H.: By the same change M may be supposed to fulfill the condition in 69.28.1 on the free rootforms. Therefore by 69.28.1 it is some C(p). The next assertion is proved the same way as that of 69.26 or derived from it. As to the last assertion, suppose that ZG(p) = G(a) for some Z E Int G. Let B be the maximal solvable subalgebra spanned by H and the ea with

t See footnote, p. 408.

410

68-75.

TITS GEOMETRIES N

B c G(p), B c C(u), a"Bc G(u). Now a"B = 6B with some 6 E G(u). Therefore a", if replaced with 6-'6, may be supposed to preserve B. Then a > 0.

N

a" E

B c G(p), hence G(u) = a"G(p)= G(p).

69.29.2 If M is parabolic subGEAlg Lie Com, then every element of Int Cleaving M invariant belongs to

a.

Proof For semisimple G this is proved in the same way as the analogous

statement in 69.29.1 for maximal M . By the principle used in 69.21.3 it is extended to arbitrary G.7 69.29.3 If M is a parabolic subgroup of G E Gru Lie Lin Com S S and a is in the normalizer of M , then Z M c M ; thus a" E A? by 69.29.2 and a E M , since the center of G is in any trunk, thus in M . Therefore M is its own normalizer. Since any Lie subgroup of G is normal in its closure, it follows that M is closed. If M is presented as the intersection of maximal parabolic M,, generating M , , then any a E ni M i normalizes all M , , hence M ; thus it belongs to M .

In the general case the statement is reduced to that about M/rad G in G/rad G, which is essentially in Gru Lie Lin Com S S . 69.29.4 The following have been proved

68.2.1 in 69.21.3. 68.2.2 in 69.25. 68.2.3 in 69.26. 68.2.4 in 69.26. 68.2.5 in 69.26 and 69.29.3. 68.4 in 69.21.3. 68.6.1 in 69.27. 68.6.2 in 69.28.1. 68.7.2 in 69.29.2-3. 68.7.3 in 69.28.2 and 69.29.3. 68.7.4 in 69.27 and 69.28.1. 68.7.5 in 69.29.1. 69.30. Bruhat's Lemma The intersection of two maximal solvable subalgebras of G E Alg Lie Corn S S contains a trunk of G. Proof Let M , , M , be maximal solvable. Since M I and M , are their own

idealizers, ad-cleaving with respect to G is possible in both M I and M,, and by

t See footnote, p. 408.

69. PROOFS OF THE STATEMENTS OF SECTION 68

41 1

its uniqueness also in M I n M 2 . By 69.11

M,nM,=A+E with A maximal ad-pure in M I n M 2 ,and E the subalgebra of ad-nilpotents in M I n M 2 .Now $ ( E M i ) = (0); thus $(E,Ml + M2)= {0}, hence dim(M,

+ M 2 )G dim G

Further

-

dim E.

+ dim E, dim M I + dim M2 = dim G + rank G,

dim(M, n M2)= dim A

+

dim(M, + M2) dim(M, n M2)=

which follows from 69.24.1. Hence dimG+rankGGdimG+dimA. Therefore dim A

2

rank G

which proves A to be a trunk of G. 69.31 Bruhat's lemma is usually formulated as follows: Proposition If G E Alg Lie Lin Com S S , M is maximal solvable sub G, His a trunk of G in M , and S is the normalizer of H in G , then

G = MSM. Proof One may assume that G is centerfree. Suppose g E G. Then by 69.30 M n g M contains a trunk of G, which by 68.1.1 can be written as niiH with m E M . Thus K1 gM 1 H. Now both M and ni-' gM are maximal solvable

subalgebras around H. By 68.3 there is some s E S with n i - I g M = SM Therefore S-lni-' g = ni I for some ml E M , g'= niSnil, which proves the assertion. 69.32. Proposition If G E Alg Lie Com S S , then any ad-nilpotent of G is contained in some semisimple subalgebra of G of rank 1. This is the content of Theorem 68.8.3. In the subalgebra F of rank 1 any ad-pure h#O spans a trunk. The representation of F by adjoint in the linear space G is conducible. Every irreducible component is spanned by weight vectors (43.1.1), on the basis of which h is diagonaljzed. Therefore h is adpure. This shows that 68.8.1 follows from 68.8.3 and from the special case of rank 1 which is obvious from 69.24.1. Proof of the Proposition. Let e # 0 be the ad-nilpotent. One may assume that C is the smallest semisimple subalgebra of G containing e.

Let N be the centralizer of e. Then ad-cleaving is possible in N. If N has a non-ad-nilpotent element, it has an ad-pure element h # 0. By 69.14 its centralizer would contain a smaller semisimple algebra containing e.

68-75.

412

TITS GEOMETRIES

Thus N is ad-nilpotent. Since $(EG, N ) = -$(G, I N ) = (0)and dim N L = codim N = dim 2G, 69.32.1

EG= N l .

Further e E N L . Thus there is some b E G with E(-h)

= e,

he = e.

69.32.2

After ad-cleaving h may be taken ad-pure. From 69.32.2 it follows that

AN^ N. Therefore h,N span a solvable algebra, $(h,N) = {0}, h E NL,and by 69.32.1 Zf=h

forsome f EG.

Now f will be replaced by somef’ such that still I f = h,

and moreover

kf’ = -f. Puttingf-f’

=u E

N, one must solve

(k + 1)u = (h + l ) J About the second member g = (h + l)f, one knows that zg = 2(k

+ 1) f = (k + 1)Zf - 2f = 0.

Thus to find the wantedf’, one must solve

(k + 1)u = g E N with respect to u EN.This can be done as soon as there is no v # 0, v E N with 69.32.3

(k + 1)v = 0.

Suppose there were such a v . One easily proves by induction that

[k,E”] =PI”,

(9

[I”, $1 = p h ~ - ~

Now 69.32.4

I’f’v

= t,v

I”-’.

69. PROOFS OF THE STATEMENTS OF SECTION 68

41 3

with scalars t, as can be shown inductively:

=fCtpv +(p

+ l)Rtpv-

i";

')tpv.

The first summand vanishes because of v E N ; thus Cv = 0, and by 69.32.3 the others are together

which proves that recursively 1,

=-I,

tp+,

=-(

P+2

)tp;

thus t , # 0 for p > 1 . Because of the nilpotency of C, this contradicts 69.32.4 and proves the assertion. 69.33. Proposition Let G E Alg Lie Com SS and H a trunk of G. Then for every maximal solvable B around H there is one maximal solvable B' around H such that B n B' equals H. Proof If H: is suitably ordered, B is spanned by H and the eOrwith u > 0. Then B' meeting B exactly in H must be spanned by H a n d the emwith u -= 0.

This indeed leads to a maximal solvable subalgebra. 69.34. Proposition Let G E Alg Lie Com SS, A? the manifold of all and go the subset of g, consisting maximal solvable subalgebras of G,B + E g, of those B that intersect B t in more than just a trunk. Then gois nowhere dense and closed in g. E 93 meeting B + exactly in H . Let E be the set of ad-nilpotents in B-. In the direct sum splitting G = B t E let p be the projection onto E. For a" E Int G let ~ ( a " be ) the determinant of the restriction ofpa" to E. Then g~ is analytic and y(1) = 1 , and therefore the set in ) that which q~ vanishes is nowhere dense in Int G. Nonvanishing ~ ( a " means pCiE = E, hence B t a"B- = G and consequently dim(Bt n CB-) = rank G, thus a"B- 6 go. Any B, E g has the form COB- with suitable loE Int G. Near a", there is

Proof Let H be a trunk within B+, and B-

+

+

414

68-75. TITS GEOMETRIES

some a" E Int G such that y ( i ) # 0; hence i B - $ go.Now B = CB- is near Bo = io B-, which shows that -go is dense in g. On the other hand, it is obvious that -go is open in 9. This proves the assertion. 69.35. Proposition Let G E Alg Lie Com SS, the manifold of ordered pairs of maximal solvable subalgebras of G, and9102the subset of those that intersect one another in more than a trunk. Theng,2 is nowhere dense and closed in g2and Int G acts transitively on g2\gO2. Remark This is Theorem 68.10. Proof It is again obvious that @2\9902 is open in g2. It is dense because by

the set [-go, {B2}1is dense in r@,{B2}1.This 69.34, for every fixed B2 E g, proves the first statement. If two elements rB,, BZ1and rB;,Bi1 of go2 are given, then by transitivity under Int G one may suppose that B, = B;.By 68.1.1 applied in B, = B; the trunks B , n B,, B; n B; may be supposed identical. But then by 69.33, B2 = B; as well. 69.37. Proposition If B is a maximal solvable subalgebra of G E Alg Lie Lin Com SS, then G / B is compact. Proof B is closed by 13.3. Let G be in ordered third dressing on some trunk

H c B such that B is spanned by Hand the eOrwith a > 0. Let D be the unitary restriction of G on the given dressing. Then G is the sum of B and D (as a real vector space), and one derives the assertion from Proposition 61.3. Remark Of course the proposition remains valid if

G E Alg Lie Lin Com

only.

70. INTRODUCTION OF INCIDENCE GEOMETRIES OF SEMISIMPLE LIE GROUPS

+

70.1 G E A, that was presented in linear (I 1)-space R with an ordered basis, also admits of a projective interpretation. I-projective space R , is usually defined as the set of linear subspaces of R,provided with the relationship of symmetrized inclusion. In this terminology a linear i-subspace S of R determines a projective ( i - 1)-subspace S, of R,. As acting in R,, G is called the projective group of R,. It acts separately in the manifold of the linear i-subspaces of R (projective ( i - 1)-subspaces of R#);however, for i = 0 and

70.

415

INTRODUCTIONOF INCIDENCE GEOMETRIES

i = I + I this action is trivial, for then these manifolds consist of one point only. For this reason in the sequel the smallest and largest elements of R , will be disregarded. Let R, be the subspace of R spanned by the first i basis vectors (1 < i < I). The trunk H of G is chosen as consisting of the diagonal matrices with coefficients w l , .. ., w , + ~(C W f = o). The stability algebra (infinitesimal algebra of the stability group) of R , is withp G i or q > i. Therefore just the ewP-,,,,with spanned by Hand the ewp-wcl, q < i c p are lacking, in other words, the e, with a = m j p j and m, < 0. This identifies the stability algebra of R , (or R f + )as G(p,).The identification extends to the belonging groups. Note that the stability group of R , cannot be N

larger that G(p,), since G(p,)is not in a larger proper subgroup of Int Cand the center of G is in If,thus in G(pf).Note, further, that from the projective point of view the center of G is eliminated, since it leaves all elements of R , fixed. G is transitive over the manifold F, of linear i-spaces sub R (projective (i - 1)-spaces sub R,). F, is a homogeneous space acted on by G. Its stability groups are the conjugates of G(p,). Disregarding its origin in R , one can consider F, as the manifold of conjugates of G(pJ or of C(p,),on which G acts by means of conjugation. By their realization as subspaces of R, the elements of F, and are related in some way that admits of an interpretation in group theoretical terms. First consider the triangular matrices of G (zeros below the diagonal), which form a maximal solvable subalgebra B of G. The R, are all nontrivial linear subspaces of R invariant under B, and B can be characterized by its property of leaving each Ri invariant. Two nontrivial linear subspaces of R one of which contains the other, are called incident, and any system of mutually incident elements is called a j a g . The system R I , R1, ..., R, is a maximal flag. G acts transitively on the manifold of maximal flags. The stability group of the maximal flag { R l , . .., R,} is just B, which is maximal solvable, and since I?again acts transitively on the manifold of maximal solvable Lie subgroups of G one can identify the manifold of maximal flags with that of the maximal solvable Lie subgroups, G/B. A pair of linear subspaces of R can be extended to a maximal flag if and only ifthey are incident. In group theory terms the criterion of incidence states that elements of F,, Fj are incident if and only if their intersection contains a maximal solvable Lie subgroup of G, in other words, if and only if it is a parabolic subgroup of G.

r,

70.2 G E D, can be discussed in an analogous way. Take G as presented in Section 16.ThentakeR,+zspannedby thefirsti-basisvectors (i= 1,2, . . ., I - 2), R2 by the first /-basis vectors, and R , by the first I - 1 and the 21th. (In the projective interpretation all of these spaces are lying on the invariant quadric

68-75.

416

TITS GEOMETRIES

f , f l + i = 0.) They are tota//y isotropic. T I ( i 2 3) is now the homogeneous space of (i - 2)-dimensional totally isotropic subspaces (i.e., lying on the quadric), and TI and T2 are the two manifolds of /-dimensional totally isotropic subspaces. The ( I - I)-dimensional spaces do not appear in this image as primordial elements; each is contained in exactly one of each of the sorts T Iand and is their intersection. To establish that the Tiare indeed orbits under G one must use Witt’s theorem? which says that any two isometric subspaces of R can be mapped onto each other by an orthogonal transformation of the whole of R. For I > i > 3, TIconsists of all totally isotropic subspaces of the same dimension i - 2 and each of them is conjugate with R,+, under G, that is, by a special orthogonal transformation, because the stability group of R i i 2 in the full orthogonal group contains orthogonal transformations with determinant -1. For TI and T2 one should note that R , is conjugate with R2 by the interchange of the /th and 21th basis vectors, which has determinant -1, and that the stability group of R , in the full orthogonal group consists of volumepreserving transformations only. The stability group of Ri is again G(p,), which leads to the same group theory interpretation ofT,asinA,.Allotherthingsalso go thesame way as in A, assoon asone takes thelibertyof interpretingthe system R , , R 2 , .. ., R, as a flag. This agrees with the naive notion of incidence of two elements except for the pair R,, R,. Now none of two /-dimensional totally isotropic subspaces of different kind can contain the other. The best they can do is to intersect in a n ( I - I)-dimensional subspace, and this is just what happens to R , , R 2 . Generally, incidence of elements of PI and has to be understood as intersecting ( I - 1)-dimensionally. After this interpretation, incidence i n group theory terms again means having a parabolic intersection. G E B, and C,can be dealt with in an analogous way. In both cases the sorts are constituted of totally isotropic subspaces of the same dimension, and incidence is the symmetrized inclusion relation, as in the case of A,.

r,

r,

70.3 For any G E Alg Lie Com S S by definition G(pi) was spanned by ordered Hand the branches e, with a = zcp,p,, such that p , 2 0. In terms of the fundamental weights, this condition on a can be re-expressed as

70.3.1

(Ti,

a ) > 0.

Now consider the irreducible linear representation of G in some space R‘, with the fundamental weight 7, as top weight and the 7,-weightvector s,Then G(p,)is just the subalgebra of g E G leaving .Y invariant up to a scalar factor;

t See, for instance, 0.T. O’Meara, fntrodiictioti to Qiicrrlrcitic F o r m , Springer, Berlin, 1963, pp. 97-99, or the more general proof in N. Bourbaki, AIgPbre, Chapter 9, S4, No. 3. Th. I .

70.

INTRODUCTION OF INCIDENCE GEOMETRIES

417

in other words, leaving x fixed if it is considered as a point x# of projective RL. This means that in the earlier examples can also be interpreted as the manifold Gx, in RL. For A, the representation ri is that in the space R 1of skew i-tensors on R . The manifold corresponding to in R' is that of i-vectors, that is, skew products of i vectors, which on the other hand just describe the i-dimensional subspaces of R. The case of D, is again somewhat different; the representations 7 r, ,7r2 act on the spinor spaces (see Section 49).

r,

70.4 The foregoing exploration leads to a general definition: Definition For G E Alg Lie Lin Com S S ,

r,the set of maximal parabolic subgroups of G (infinitesimally generated by the maximal parabolic sub-algebras of C ) , endowed with the incidence relation ea,b E incident iff a n b is parabolic>,

r

is called the incidence geometry of G. r i s considered to be acted on by G by conjugation. Its transitivity classes are the sorts of the incidence geometry. 70.5.1. Definition Let H be an ordered trunk of the above G. r o ( H ) ,or T ofor short, is the restriction of r t o the subset of a E r s u c h that Hca. r(H,p),or r ( p ) for short, for p E W + + ( G , H )is , the transitivity class of containing G(p). Its intersection with is denoted by ro(H,p)or r O ( p ) .

r

r0

70.5.2. Proposition Int G-equivalent a, b alent.

E

r O ( p ) are Int(G,H)-equiv-

iia = b for some u E G . Both H and H' = iiH are trunks of b. The conjugacy of trunks provides a uo E b with iioH' = H. Then C0C E Int(G,H) and iioiia = frob = b.

Proof

Remark The proposition follows also from 69.28. I and 69.29.1.

r=

r,

70.5.3. Proposition U p E W + + r ( pwhere ), r(p) # for p # u. On Fo the action of Int G is like Int(G,H)/A, isomorphic with Int W * . Its transitivity classes in T oare the Y0(p). This is only a reinterpretation of known facts. Information on incidence properties in r c a n already be obtained to a high degree in To,which is finite. This explains the importance of in the study of As an example:

r.

ro

418

68-75.

TITS GEOMETRIES

70.5.4. Proposition If a,b E r ( p ) are incident, they are equal.

By the action of Int G it may be assumed that a n b contains the maximal solvable subgroup generated by H and the e, with a > 0 . Then a = G(p), b = G ( p ) ; hence a = b. Proof

ro

70.6 Any a E equals SG(p) for some p E Wtt and S E Int(G, H). As usual, Swill also be considered as an element of Int W*. Henceforth write n, instead of n1 if p = pi. According to 70.3.1, G(p) is spanned by Hand the e, with (n,, a ) 2 0.

Therefore, SG(p)is spanned by Hand the eSawith (n,, a) 2 0,

in other words, by H a n d the e, with (Sn,, a) 2 0. This correspondence suggests the following : 70.6.1. Definition n(p)= (Int W * ) x p for p E W++,

n = u, n(p). w(SG(p))= Sn,

if S

E

Int(G,H);

so w maps YO(p)one-to-one onto I7 (p). A, p E I7 are called incident if (A, a)&, a ) > 0

for all a

E

W*.

Obviously I7 consists of all integral elements that are equivalent to fundamental weights. Clearly: 70.6.2. Proposition If a E spanned by Hand the e, with

r0,then

the infinitesimal algebra of a is

(wa, a ) 2 0.

The last part of the definition above is justified by the following: 70.6.3. Proposition a,b E T oare incident if and only if wa, wb are so. Proof Mcontaining a maximal solvable subalgebra around Hmeans that for any CL E W* either e,, or e-y E M . Therefore, a n b to be parabolic means that

either (wa,y) > 0 A (wb,y) 2 0

or

(wa, y ) G O A (wb, y ) Q 0.

70.

INTRODUCTION OF INCIDENCEGEOMETRIES

419

In other words, (wa,y)(wb,y) 2 0, which proves the assertion. Evidently : 70.6.4. Proposition By w the action of Int(G,H) in T ois translated into that of Int W* in 17. 70.7. Proposition If F c naGFa is parabolic.

r consists of mutually incident elements, then

Proof By Proposition 70.5.4 F is finite. The proof is inductive on the number of elements of F. Take the statement for granted for F' = F\{c} (some c E F). There are maximal solvable subgroups B, B' with B c flasF,a,and B' c c. Bruhat's lemma provides a trunk H' c B n B ' ; thus H' c flaEFa. So one may suppose F c The mutual incidence in F then means

ro.

(A, a)&, a ) > 0

for all a

E

W*

and

A, p E wF;

thus, if a E W* is given, either (A,a) 0 for all A E wF or ( & a ) G 0 for all of them. Consequently, for any a, either e, or ePoris an infinitesimal element of all a E F, which shows by 69.28.3 the existence of a maximal solvable Lie subgroup in their intersection. 70.8. Definition The intersections of closed halfspaces bounded by a L ( a E W*) are called parts (of H:). If V C W* such that V u ( - V ) = W*, then the intersection of the halfspaces ?&$,a) > 0 with a E V is called a cell. An

edge is a one-dimensional cell.

Clearly every closed chamber is a cell, and any cell is the intersection of closed chambers. Parts are convex. An i-dimensional part is unambiguously built up from i-dimensional cells, which are minimal i-dimensional parts. By the definition of fundamental weight, the edges contained in the closed dominant chamber are theimagesof yT20(T.rrp)( p E W++). Thanks totheaction of Int W*, it follows that the edges are the sets of weakly positive multiples of elements of 17. Proposition For A,p E 17the following statements are equivalent: A,p are incident; h , p are in a common closed chamber; A,p are in a common cell. E r t h e corariant r(a,b)of a,b denotes the set of all c E F that are G-fixed when a and b are G-fixed. The stability groups of a,b,c are just their adjoint images (see 68.7.2). a,b are simultaneously invariant under precisely those P for which x E a n b ; the invariance of c under the same 2 means that c =I a n b. So:

70.9. Definition For a,b

420

68-75.

TITS GEOMETRIES

Proposition r ( a , b )is the set of c = a n b, c E Clearly, if a, b E then r ( a ,6 ) c

ro,

ro.

r.

70.10. Definition The smallest part containing A,p is denoted by P(A,p). If A,p E then 17(A,p)means P(A,p) n

n.

n,

Proposition n ( w a , w b ) = wr(a,b) if a,b E

ro.

Proof Translated by w, c 2 a n b states that AYEw*{[(wa, y) 2 0 A (wb, y) > 01 + (w,y ) 2 01.

In other words, wc is found in every part that contains wa and wb. 70.11. Proposition If G E Alg Lie Corn SSS and there is an a E W* with ( f l , u ) # 0 for i = 1,2.

C1,t2 E Zf:\{O},

then

Proof The set of a E W* with (gi,a) = 0 is called W, (= - W,). Suppose the statement is wrong. Then WI u w, = w*.

W, # W* because the inner product is nondegenerate on H*. Take

If then

+ a, E w ;

thus either a,

+ a, = 0

aI

+ a, E W,

or

(hence a I E W,, which is impossible) (hence a I E W,, which is impossible).

Thus aI

+ a2 E W1\W2.

For V c W*'the linear span of the ea with a has been proved that

[6(Wl\ W2h a(W,)l=

+

E

V is denoted by 8(V ) . Then it

Wl\ W2).

The space 6(W,\ W,) is idealized by H a(W,), and so is the Lie algebra D generated by 6( W\, W,). Since D c H + 8(W,) # G, it is a nontrivial proper ideal of G, which contradicts the simplicity of G.

70.

INTRODUCTION OF INCIDENCE GEOMETRIES

42 1

70.12. Proposition If X,p E 17 are incident, then (h,p) > 0. In the case of simple G this inequality is strengthened to (X,p) >O. Proof According to 21. I .9,

(A, p) = ,f;,*

(k 4 ( p ,

4

9

where all summands are >, 0 if X,p are incident. (h,p) = 0 is possible only if all summands vanish, but according to 70.1 1 this is forbidden by simplicity. 70.13. Definition For x J(x). Theorem

ro.

r t h e set of elements incident with x is called

Suppose that G is simple. If J(a) n J(b) # 0, then there is a c

Proof

E

EJ(a)

with J ( a ) n J(b) = J ( c ) .

n J(b)

By Bruhat’slemma,a n bcontainsatrunk.Thusa,bmaybesupposedin

70.13.1

cE

J ( u ) n J(b) n To,

fulfilling 70.13.2

J(U)

n J(b) n T o= J(c),

also fulfills

J ( a ) n J(b) = J ( c ) .

70.13.3

Indeed, if d E J ( a ) n J(b), then both a n c and b n d are parabolic and therefore a n b n c n dcontains a trunk H’. Now H = XH’ by means of some .Y E a n b n c; thus

Xu = a,

Xb = 6 ,

Xc

= c,

Xd = d‘ E To,

and again d’ E J ( u ) n J ( b ) ;

hence by 70.13.2 d‘ E J ( c ) ,thus

d E JW, which proves 70.13.3. At the same time the following has been proved :

422

68-75.

TITS GEOMETRIES

70.1 3.4 If a,b E To,J(u) n J(b) # 0, then J(u) n J(b) n T o# 0. A c that fulfills 70.13.1?2 will now be constructed. Put A = wa, p = wb. Then

w(J(a) n J(b) n To) is the set N of elements v of 17 such that

(A, a)(v,a ) 2 0 A (p, a)(v,a) 2 0 Note that for v

E

for all a

E

W*

N,

(A, a)&, a) < 0 -+ (v, a ) = 0. Let F be the subgroup of Int W* generated by the reflections S,

with (A, a) = (p, a) = 0.

Then A, p are invariant under F, as is N . For v E N put v'

=

c sv.

SEF

Then Sv' = v' for S

(A,

E F, particularly

q ) = (p, a ) = 0 -+

for S, with (A, a ) = (JL, a) = 0; thus

(v', a ) = 0

for a E W*.

Further

(A, a ) > 0 v (p,a ) > 0 -+ (v', a) 2 0, for under this condition (v,a) 2 0 for all v E N and N is invariant under F. Finally (A, 4(p, 4 < 0 -+ (v', 4 = 0, since this is true with all v E N instead of v'. According to 70.12, (A,v) > 0, (p,v)> 0, and, because of the F-invariance of A, p, also (A, v') > 0,(p,v') > 0. Thus v' # 0. Now define D as the set of 5 E H: such that for all a E W*

(A, 4 = (p, a) = 0 -+ (5, a) = 0,

(A, and

a)2 0 A (p, a ) 2 0 -+ (5, a) 2 0,

(A,4(p,

(D)

4 < 0 (5, a ) = 0. +

Then D has been proved to contain an element v' # 0. D is a positive-dimensional cell. It contains an edge and thus some

~ ~ D n l 7 c N .

70.

INTRODUCTION OF INCIDENCEGEOMETRIES

To use IC-I K as c, one must verify the conditions on 70.13.1-2; that is, K

incident with A,p and with every v

The first two assertions are evident, since K (K,

a ) ( v ,a ) 2 0

E

for all a

E

K

423

corresponding to

N.

N . The third means E

W*,

v E N,

which needs verifying only for ( ~ , a>) 0. Now in this case, by the definition of D,(A, a ) > 0 or (p,a ) > 0; hence, according to the definition of N , (v, a ) > 0 and therefore ( K , Q ) (v,a) 2 0. By this result the existence of c is granted. 70.14 Every maximal solvable Lie subgroup of G@) contains the radical rad G(p) of G(p), which is generated by rad G @ ) .The radical of C @ )is spanned by some h E H with h#O and u(h) = 0 for all u E W i '-\{p} and by all errthat involve p with a positive coefficient. rad G(p) can be used to decide whether some b E T ois incident with C(p). More specifically: 70.14.1. Proposition b E T is incident with a E riff b contains rad a. The necessity of this condition is evident. Its sufficiency need be proved only for H c a n b and a = G(p). It is then an immediate consequence of the following: 70.14.2. Proposition If b E T ocontains rad C(p),then b is equivalent to some G(a) by means of an 2 E Int(G,H) with x E G(p). Note that'G(p), G ( o ) are always incident. Proof By 20.15 the e, and h, in which a does not involve p span a semisimple

subalgebra GI, with the span ofthe h,(=kernel of r,,)asatrunkH,. Obviously GI c C(p). If C , c 6, then obviously b = C @ ) . If GI b, then b n GI is a parabolic subalgebra of G , ; under a n element of Int(G,,H,) every free rootform (see 69.23) of b n GI may be assumed positive. However, Int(G,,H,) is naturally included in Int(G,H) and its elements leave rad C ( p ) invariant. Therefore the assumption entails b = G(a) for some E W . + +u, # p. (3

70.14.3 Another immediateconsequence of 70.14.2 is stated i n the following: Proposition If b E T ois incident with G(p), then b is equivalent to some G ( o )by means of an element of Int(G,H) that does not change G ( p ) . This proposition allows one to assume incident a,b in the more convenient form G(p), G(o).

424

68-75.

TITS GEOMETRIES

70.14.4 By the same method the foregoing can be strengthened to obtain the following: Proposition If a = G(p), b E Fo(u), c E Yo(,), p between u and T in W++ and a incident with both b and c, then by an element of a leaving H invariant b and c can be carried over into G(u) and G(T). It follows from the fact that rootforms that do not involve p cannot involve both u and T .

70.15. Definition A chain of lengthp in r i s a sequence ofp + 1 elements of in which every element is incident with its successor. A chain is called irreducible if all its elements are different.

r

Theorem If a, b E r a n d a pair ofdifferent p, u E W t t are given, then a chain cO,clr . .., ck,cktlcan be found with co = a, cktl= b, c j E r ( p ) u r(u) for j = 1, ..., k.

ro.

It may be assumed that a, b are in Then in 17 a sequence Ao, A l , . . ., &,Ak+, must be found such that A, = wa,A,, = ti.6, and Aj is incident with h j + I .In any case there is a sequence of closed chambers Co, . . ., ck+lsuch that dim(Cj n Cj+l)= I- 1, wa E Co, and wb E ck+l. Now, since C j n cjil contains one element of each sort but one of (see 70.6.3 and 70.8), Cj n Cj+l n (17(p) u n(u)) # 0, and the only thing one has to do is to choose Aj+, in this set.

Proof

n

r

r,

70.16. Definition r m o d a =J(a)\,{a} for a E with the notions of and J as defined in 70.4 and 70.13. mod a is endowed with the incidence structure that it bears as a subset of

r r.

r

Theorem If a E r ( p ) , then mod a is isomorphic to the incidence geometry of G(p)/rad G(p), the graph of which arises from that of G bydeletingpand its bonds with other elements. More precisely, if, for the sake of convenience, a = G ( p ) is assumed: Let CJI be the canonical mapping of G(p) onto G(p)/rad G(p). Then 9 = YbErmodo~(b n G(p)) maps mod a isomorphically onto the incidence geometry ofdG(p)).

r

Proof Since y(G(p)) is semisimple, d exists. By 70.14. I r m o d a is the set of maximal parabolic subgroups of G around rad G(p). The elements of mod a contain maximal solvable Lie subgroups G ( p ) (necessarily containing rad G(p)). Since CJI maps the maximal solvable Lie subgroups of G(p) onto those of G/rad G(p), 9 preserves the parabolicity of the elements of mod a ; clearly it maps mod a onto d (to show that 0 is onto, put an element of d in its

r

r

r

70.

INTRODUCTION OF INCIDENCE GEOMETRIES

425

canonical form). The mapping 6 preserves incidence, for cp maps maximal solvable subgroups onto maximal solvable subgroups. Similarly, it is evident that the incidence of 6b,& implies that of b,c. Finally, let 66 = 6c. Then by the foregoing a,b,c are mutually incident, so by 70.7 a n b n c is parabolic and may be assumed to contain H and all emwith ct E W+. This, however, means that a,b,c are some G(p), G(u), G(T).But it is then evident that 6 b = 6 c implies b = c, which completes the proof that 6 is isomorphic. The remainder of the statement is evident. 70.17 If G splits directly into GI, G,,then the maximal solvable Lie subgroups of G split correspondingly into maximal solvable Lie subgroups of GI and G2, and a maximal parabolic subgroup of G contains GI or G,. Those containing G, form a subset of (i # j ; i = 1 , 2 ; j = 1,2). Clearly every element of is incident with every element of Moreover, Ti, endowed with the incidence structure induced by is isomorphic to the incidence geometry of G i . This suggests a definition of direct sum of incidence geometries which will be substantiated later. Meanwhile is called simpre if G is simple.

r, r r,

r,

r,.

r

70.18 As an immediate consequence of 70.1647, one gets the following: Theorem If, in the graph of G, p lies between u and T , then every element of r ( u ) n J ( G @ ) ) is incident with every element of r(T)n J(G@)). 70.19 The following is a refinement of the statement of 70.13.

r be nonincident, and J(a) n J ( b ) n r@) # 0.

Theorem Let I' be simple, let a, b E

Then there is d E J ( a ) n J(b) such that 70.19.1

J ( d ) n r ( p ) = J ( u ) n J(b) n Q).

G E A,. Then r may be identified with projective I-space and a,b with elements of it. Foranyp onecan takefor d either the intersection or the span of a, b (if it exists in i.e., has linear dimension#O, # / + 1). Remark To illustrate this statement take

r,

E r(u), b E r(7). Since a, b are nonincident, and because of 70.18, p does not separate u and T in the graph of G. According to 70.13, there is a c E J(a) n J(b) with

Proof Suppose that a

70.19.2

J(a) n J(b) c J(c).

Suppose that c E r ( K ) . Again K does not separate U , T . Moreover, K # u,K # T .

426

68-75.

TITS GEOMETRIES

If K = p or K separates p and u (and thus p and T ) , then by 70.18 every is incident with a and b. Therefore element of J ( c ) n r(p) 70.1 9.3

J(c) n F(p) = J(u) n J(b).

By combining 70.19.2-3 one can satisfy 70.19.1 by putting d = c. If not, W++\{K} has a component containing p, U , T . Using the terminology of Theorem 70.16 with K instead of p, one considers mod c which by 6 is mapped onto d. Now corresponding to the breach K in W++\{K}, d breaks into d and d2,where d belongs to the component containing p, u,T ; therefore 6 maps F(p) n J(c),a, b into d ; furthermore, d ,A , are elementwise incident. Now by induction the theorem may be taken for granted with respect to A , . This means the existence of d' E d I with

r

d' E .I(& n J(6(b)) )) and

J(d') n d ,@)

= J(6(a)) n J(6(b)) n d

The contribution of d2being trivial because of the elementwise incidence of d A,, one may replace d by d in this equation. Finally, by applying 6-' one gets a d which fulfills 70.19.1 (again using 70.19.2).

,,

,

70.20. The Special Case of rank 2 Gis supposed to be of rank 2. The primitive rootforms on the ordered trunk H are p,a.

(sps2= 1, where for 70.20.1

The number of chambers is 2m. According to adjacency, they are put into a cyclic order that also applies to the elements of I7 as vectors in the plane. In this cyclic order elements of l7@) and n(u) alternate with one another. Neighbors are incident. If A, p, v E 17 and h + p # 0, then v is called between h and p if v = sh + rp for suitable positive s,t . By w-l the cyclic order and the betweenness are transferred to ro. In an irreducible chain within rothe members follow each other according to the cyclic order (or its inverse); if the length of the chain is < m, then its intermediate members are between the extremities. 70.20.2. Proposition If c is between a and b in

ro,then a n b c c.

Proof Let ea E u n b. Then (wa, a) 2 0,(wb,a) > 0. Thus, if c between a and b,

(wc,a) 2 0, thus ea E c.

71.

427

AN AXIOMATIC APPROACH

70.20.3. Proposition If the extremities of an irreducible chain of length < m are in To, then the entire chain is in To.

ro,

Proof Let a,, . .., a, constitute an irreducible chain, ao,ap E p < m. The proof proceeds by induction on p . Both a, and up-, n up contain a maximal solvable Lie subgroup of G. Therefore by Bruhat's lemma there is a trunk H' c a, n ap-, n ap. Both H and H' are in a, n a,. Thus XH' = H by means of the inductive argument can be applied to some x E a, n a,. Since Xu,-, E To, the chain ofRa, ( i = 0, . . . , p - 1) to the effect that Xu, E Tofor i = 0, . . . , p - 1. Since x E a,, this extends to 2a, E for i = 0, . . .,p . The supposition p < m guarantees that Xul is between Za, = a, and Zap= a, for 0 < i < p ; thus by 70.20.2, x E a, n up c Xu,, hence x E a,, a, = 2a, E To, which proves the assertion.

ro

70.20.4. Proposition The intersection of an irreducible chain of length G rn contains a trunk of G. Proof Let ao,a,, .. ., a, be the chain. Then there is a trunk H' in a, n up-, n up. Nowp - 1 < m ; thus by applying 70.20.3 with H' instead of Hone gets H' c a,

f o r O < i < p - l aswell. 70.20.5. Theorem If G is of rank 2, then any pair of elements of r c a n be joined by a chain of length Q m (see 70.20.1). If, moreover, there is a chain of length < m to join distinct a, b E then there is exactly one irreducible chain of length < m to join them.

r,

Proof The first assertion follows from the fact that any pair a, b E T may be supposed within To.Let a, b be joined by an irreducible chain C of length Gm.

ro.

Then by 70.20.4 the entire chain C may be supposed within Let a, b (now E To) be joined once again by a chain C' of length < m. Then by thinning out, C' may be assumed irreducible, hence by 70.20.3 contained in To. Then either C and C' are equal or C u C' = To.The alternative, however, leads to a contradiction, for the lengths of C and C' are Q m and < m, whereas Tohas 2m elements. This proves the theorem. 71. AN AXIOMATIC APPROACH TO INCIDENCE GEOMETRIES OF SEMISIMPLE LIE GROUPS 71 . l . Definitions 71.1.1 An incidence geometry consists of a set and a reflexive symmetric binary relation called incidence.

428

68-75.

TITS GEOMETRIES

r

is called the direct sum of the incidence 71.1.2 An incidence geometry geometries rl,. . ., if as a set I' is the disjoint union of TI, . . ., the incidence relation on is the restriction of that on are elementwise incident for i # j . In this situation the are called direct summands of

r,

r,,

r,

r,,r,

r,

r,

r.

r,

r

71.1.3 If F i s an incidence geometry and a is an element of then mod a is the incidence geometry consisting of the set of elements incident with a, except a itself, and of the restriction of the incidence relation on F. 71.1.4 Any graph to be mentioned is supposed to consist of a nonnegative finite number of dots and a finite number of mutual bonds which may be multiple and directed. A splitting of a graph into disjoint unconnected subgraphs is called direct. A graph gets reduced with respect to one of its dots by omitting that dot and its bonds with other dots.

r,

71.1.5 An incidencegeometry on agraph consists of an incidence geometry a graph K , and a mapping of the set of dots of K onto a set of subsets of such that the images of the dots are pairwise disjoint, the union of the images of the dots is incident elements of the image of a dot coincide. The notion of isomorphism of incidence geometries on graphs is understood in the obvious way. The number of dots is called the rank of the geometry.

r

r,

r rl,

71.1.6 An incidence geometry on a graph is called the direct sum of the incidence geometries on graphs . .., I', if, as an incidence geometry, rsplits directly into the incidence geometries rl,.. ., and this splitting is induced by one of the underlying graph into subgraphs belonging canonically to the In this situation the are called direct summands of If I' has no nontrivial proper direct summands, it is called simple.

r,

r,,

r.

r

r,. r,

71.1.7 If is an incidence geometry on a graph and a is an element of then mod a is interpreted in a canonical way as an incidence geometry on the reduction of the graph of with respect to the dot to which a belongs. More precisely, I' mod a is, by restriction of the incidence relation and intersection of the images of the dots, an incidence geometry on the reduction of the graph of with respect to the dot to whose image a belongs.

r

r

r

71.2. Definition A system Z o f incidence geometries on graphs is called a Tits geometry if it fulfills the following conditions:

71.

AN AXIOMATIC APPROACH

429

71.2.0 In every element of Z the image of each dot is nonvoid. 71.2.1 For every

one of

r itself.

71.2.2

Every direct summand of F E Z belongs to Z.

71.2.3 If

of

r E Z a n y direct splitting of the underlying graph induces

r,,r,.

71.2.4 If

r,,r, E Zare disjoint, then there is a r E Z, which is the direct sum

r E Z a n d a is an element of r,then r mod a E Z.

71 -2.5 For any

I ‘ E Z,ifa,bareelementsofr,andifr(p),r(o)aretheimages of different dots p, u of then there is a number k and a sequence co, .. ., ck of elements of such that a = co, c, = b, c,-, incident with c, (i = 1, ..., k ) , and C, E F(p) u for i = 1, . .., k - I .

r,

r r(o)

71.2.6 For any simple F E Z, if J ( x ) means the set of elements of F incident and if a, b are nonincident elements and r ( p ) is the with the element x of image of a dot of then

r,

r,

if

J ( a ) n J ( b ) n r ( p ) # 0,

then

J ( c ) n r ( p ) = J ( a ) n J(b) n r ( p )

for some c E J ( a ) n J(b).

71.2.7

r E Z a n d r’isomorphic to r implies r’ E Z.

71.3 An incidence geometry of a complex semisimple Lie group can in an

obvious way be interpreted as an incidence geometry on a graph. Then some major results of Section 70 can be summarized in the following theorem: Theorem The system of geometries isomorphic to the incidence geometries

of complex semisimple Lie groups is a Tits geometry. In particular, the validity of 71.2.1-3 is granted by 70.17, the validity of 71.2.4, 71.2.5, 71.2.6 by 70.16, 70.15, 70.19. 71.4 A Tits geometry is to a high degree determined by its members of rank 2

If these members are prescribed, then a great many properties of any member of the system can be derived from its graph. This will be illustrated in 71.6

430

68-75. TITS GEOMETRIES

In particular the Tits geometry of Theorem 71.3 is essentially obtained if only graphs of semisimple groups are admitted and if the former correspondence between graphs of Lie groups and incidence geometries for the case of those of rank 2 only is postulated. However, this will not be proved here. 71.5 The property of incidence geometries of Lie groups of rank 2 stated in Theorem 70.20.5 is generalized by the following definition. Definition of generalized polygons: A generalized m-gon I' is an incidence geometry on a graph of rank 2 such that any pair of elements of can be joined by a chain of length Gm, if two elements of can be joined by a chain of length cm, then there is exactly one irreducible chain of length <m to join them. (The notions of chain and irreducible chain are understood as in 70.15.) The most convenient kind of graph for a generalized m-gon is a pair of dots with a (m- 2)fold bond. Then, however, to keep in line with the results of 70.2 one must indicate G2not by a threefold but rather by a fourfold bond.

r

r

71.6 In the remainder of this section incidence geometries to be analyzed are supposed to have been taken from a fixed Tits geometry with the special property that all graphs involved are graphs of semisimple groups, the geometries of rank 2 are generalized polygons (2-gons, 3-gons, 4-gons, and 6-gons for the graph of A, A,, A,, B2, and G2). A few examples will be discussed. In a given geometry I' between given elements a shortest chain is constructed, starting with some chain granted by 71.2.5 and shortening it by the use of the other postulates : If in a chain

+

...bat

...

r

b,a,c belong to r(u), r ( p ) , T(T),respectively, then in mod a, which by induction may be supposed to be better known than 1',there is a chain from b to c that can be inserted instead of bac. If u , are ~ separated by p in the graph of I', then by 71.2.1-4 there is incidence between b and c, so that a may be simply omitted. By such operations one may succeed in shortening the original chain. The dots of the graph will be numbered 1, . .., 1 in the usual way if 1 is the rank of The elements of the ith sort will be indicated by i, i', i", and the like. However, to avoid distinguishing marks the same sign i will be used even for different elements of the sort i occurring in the same chain when no confusion is likely to occur.

r.

71.

AN AXIOMATJCAPPROACH

43 1

ExampIes

71.6.1 A,. Between any pair of elements there is a chain of length ~ 3 . Therefore if 1 , l ' are given: 121', which is unique. In the same way if 2,2' are given, 212'. Interpreting the dots 1,2 respectively as the sets of points or lines, the incidence axioms of the projective plane are fulfilled. Pis a plane projective geometry. 71.6.2 A,. The elements of sort 1,2,3, . .. are called points, lines, planes, .... By induction on I it will be shown that two points can be joined by a line. In any case, there is a chain 121'2'12 * * 1.

One considers I" = r m o d l ' with the graph of A,-,. In r' the roles ofpoints, lines, . . . are played by elements of sort 2,3, . ... Therefore the neighbors 2,2' of 1' can be joined by 232'. Substitution changes the given chain into 1232'12

*

-

*

1,

in which the 2,2' between 1 and 3 may be omitted by 71.2.4 and 71.2.1. The resulting chain is 1312 * * * 1.

r

In mod 3 one considers the "lower" direct summand, which is a projective plane, in which points can be joined by a line. This allows one to replace 3 with 2. The new chain 1212 ... 1 is two links shorter than the original. This reduction can be repeated until 121, which shows the existence of a line joining two given points. Its uniqueness for distinct points is shown as follows: Let I21 '2'1 be a (closed) chain. By 71 -2.6applied to 1 , l ' instead of a,b, there is an i incident with 1,2,1', 2'. In mod i this argument can be repeated until one arrives at a geometry of rank 2 in which such a chain is excluded by assumption. This proves the uniqueness ofjoining points in Another property is that a line and a plane, if incident with two different points, are incident with each other. Indeed, if line 2'and plane 3 are not incident, by 71.2.6 there is an i incident with 2,3, and the two points. Again, arguing by r mod i, one arrives at a 2' or 3', incident with the given 2 and 3. Then 2 = 2' or 3 = 3', both of which are contradictory.

r

r.

432

68-75.

TITS GEOMETRIES

Three different lines intersecting in three different points are in a plane: one is given the closed chain

where the stroke is used to denote incidence. As before, one may replace some 1 by 3, /3\

2

h

1-2-1

and thus omit the two upper 2’s,

3 1’ 2’‘

1‘

From the property already proved, it follows that 2,3 are also incident, which shows that 3 is incident with all given 2’s. Continuing this procedure, one can prove that F fulfills all incidence axioms of projective space. 71.6.3 B,(C,). Given nonincident 1,2, there is a chain of length G 4 joining them; therefore 1212, which is even unique. Given that 1,1 may by chance be incident or possess a unique chain 121(joinable by a line). In any case, there is a chain 12121 between them. This is the geometry on a 3-quadric in projective 4-space or, equivalently, the symplecticgeometry of projective 3-space. 71.6.4 B&). and planes:

In the symplectic interpretation 1,2,3 are the points, lines,

._.-. 1 2 3

(1) For any 1,2 there is a chain 1212, which is unique if there is no chain 132. (2) A chain 132 is unique for nonincident 1,2. (3) For any 1,3 there is a chain 1323,

71.

AN AXIOMATIC APPROACH

433

which is unique for nonincident 1,3. (4) For any 2,3 there is a chain

2313, which is unique if there is no chain 213. (5) A chain 213 can be completed to

2-1-3

\/\/ 3-2

within the geometry reduced with respect to 1 . The chain 213 is unique if 2,3 are nonincident and so is the completion. (6) A chain 121 is unique if 1 , l are different. (7) A chain 2312 can be completed to

3-1

/\ /\

2

1-3 If there is no chain 212, it is unique and so is the completion. (8) A chain 212 can be completed to

2-1-2

\3-2-3 /I\/

It is unique if there is no chain 232; its completion is not. (9) A chain 232 is unique if 2,2 are different. (10) A chain 313 can be completed to

3-1-3

/\I2-3-2

It is unique if there is no chain 323; its completion is not. (1 1) A chain 323 is unique if 3,3 are different. (12) For any 1 , l there are chains 12121 which are never unique; likewise for 2,2 and 3,3. The facts from (5) onward, and (2), areeasilyproved. The existence of chains (l), (3), and (4) will be proved by an explicit construction; in this procedure elements with respect to which mod-reduction took place are underlined.

434

68-75.

TITS GEOMETRIES

(3) One starts with a chain 132323

23 of unknown length:

*

-3

1323232 - *

*

1321232 --

* *

-

3

*.* 3 13132 -

123232 -

3

* 12132 -

3

* *

13232

3

by which the chain has become two links shorter. This shows the existence of 1323 for given 1,3. (1) One starts with 13232 * 2 which by (3) can be shortened to 13232; hence, 13232 13212 -

--

1312 1212 Thisshows theexistenceof 1212. 2313131 (4) -

-2312131 -

23232131 -

*

3

*

3 3

* *

2323131 -

* * *

3

2123131 -

* * *

3

213131 -

* *

212131 -

3

2123231 213231 -

3 3

* * .

* * *

3

2323231 -

* * *

3

2321231 --

* * *

3

23131

*

a

-3

which shows the existence of 2313. As an example of a uniqueness proof, that for 1323 is given :

7 1.

AN AXIOMATIC APPROACH

435

(In the last step 71.2.6 is used.) Reduction mod the lower 1 shows two chains 2323 in a B,-geometry, which is forbidden.

71.6.5 F,:

._.-._. abed

Given a,d, there is a chain adad. In the proof the properties of B, are involved if reducing mod a or mod d. One may start with

adcdcd

d

*

and suppose that at least two pairs cd are present (if there is only one, it may be doubled).

.. d

adcdcd -

acbcdcd d .d acbdcd -

-

acbcacd acbcad adad

-

*

*

acdcad - a

-- d -- d

-

* *

d

d

9

If t h m are still pairs cd present, one continues as follows: adadcd .. d adacbcd - ** d adacbd * * d

adcdcbd - -- d adcdbd - - - d adcdcd - . . d *

*

which by the first procedure and by induction leads to adad.

436

68-75. TITS GEOMETRIES

71.7 Although interesting in itself, this haphazard search for shortest chains and uniqueness proofs is not satisfactory. A more algorithmic method would be desirable, though in the general case its existence might be doubted. If, however, one restricts oneself to incidence geometries of semisimple groups, group theory methods are available to solve these kinds of problems. They will be developed in the next sections. 72. COVARIANTS OF PAIRS OF ELEMENTS IN INCIDENCE GEOMETRIES OF SEMISIMPLE GROUPS

G is supposed to be semisimple complex. Other notations are taken from Section 70. 72.1 F(a,b),as defined in 70.9,is the covariant of a,b E I‘. By choosing the trunk H within a n 6, F(a,b) has been related to ~ ( w uwb) , =P(wa,wb) n I7 by 70.10.P(h,p) is the smallest part of H: containing X and p. Since two trunks of Q n b are equivalent under inner automorphisms of G leaving a and b invariant, the metric properties of 17(wa, wb) are covariant properties of Q and b. U(X,p) will be discussed for any A,p E 17. 72.2 For a E W* the set of 4 E H: with (5,a) = 0 is the hyperplane a l . The line-interval [A,p] is contained in the intersection N of all hyperplanes a’passing through X and p. Definition dim N - 1 is called the rank ofthepair h,p E 17. Proposition If the cell C contains several points of [h,p],then dim C > dim N = rank ‘X,pl + 1. Indeed, such a cell contains a subinterval of A[ ;], its linear span contains h and p, is the intersection of some hyperplanes aL containing h and p, and therefore contains N .

72.3 Let (,, . .., .fp-l E [A,p] be the consecutive intersection points distinct from X,p of [X,p] with cells of dimension G rank rX,pl, counted from h to p. In addition, put lo= A, 5, = p. Let Cibe the smallest cell containing 5,and put

D i = C i n17

( i = O ! l , ...,p).

Note that Do = 01,

Dp = {PI.

Every part containing h,p contains [A,p],thus ti,and therefore the smallest cell in which El lies. Hence: Proposition

D , c 17(A,p) for i = 0, 1, . . ., p .

72.

COVARIANTS OF PAIRS OF ELEMENTS

437

Note that some D,may be void. Then 4, = 0 and [A,,] contains the origin; thus -p is a positive multiple of A. F o r some S E Int W * , SA is a fundamental weight 7r. Now S(-p) is also dominant, hence a fundamental weight, hence S(-p) = Sh. Thus p = -A. Moreover, rank ‘A,’ = 0, p = 2, D I = 0. 72.4 The elements of (5,) u D i are not separated by a hyperplane ’a same is true of the elements ti..,and tiand of the elements of ( f i - 1 ) u Di-1.

( a E W * ) .The

Proposition The elements of Di-, u D,are mutually incident.

72.5. Proposition Proof

Let v

E

For 0 Q i <j < k G p : D , n Dk c D j .

D , n Dk. Thus for all a (v, a)(5i,a )

0,

W*

E (

~

c9 O ( t k ,

a)

0.

I f ( v , ~#) 0, then ( f i , a ) ,( f k r a have ) the same sign in the weak sense, thus them still has the same sign. Hence

( f j , a ) between

(v,

a)(tj,a ) 2 0

for all

a

E

W*,

which proves the assertion. 72.6. Notation E, = N - I D i . Notethatifrank ‘ A , p 1 = O then

E=

E = ‘E0,..., Epl.

‘{H-’ A},

0, (1i*-’p}’.

u El are mutually incident.

Clearly the elements of

E,

=

{a},

E,, = {b}.

Definition A sequence ‘E,, . . ., Epl of finite subsets of F i s called a chain if the elements of u Ei are mutually incident. Two chains ‘E0, ..., E p l , ‘E:, . . ., E;I are called isomorphic by means of g, if g, maps U Eione-to-one onto U ET, such that Ei is mapped onto ET, and for any p, if x E r ( p ) , then y x E T(p). Clearly, according to 72.4, the €defined before is a chain, which u p to Int Gequivalence does not depend on the choice of the trunk in a n b.

72.7. Theorem The pairs ‘a,bl and ‘a*,b*l are equivalent under Int G if and only if Edetermined by ‘a,bl and E* determined by ‘a*,b*l (both with respect to some trunks) are isomorphic. Proof “Only if” being obvious, one supposes that E, E* are isomorphic; hence p == q. Since a,u* belong to the same sort, they are equivalent under Int G; thus

438

68-75.

TITS GEOMETRIES

they may be supposed equal. Both a n b and a n b* contain trunks of G, which areconjugatein a. Therefore, withoutchanging a = a*, one may arrange to have a common trunk Hof G within a n band a n b*. This trunk is used in the sequel, E, E* may be supposed to be constructed on I€.

Eo = {a} = {a*} = E,*. Since the elements of Eo u El as well as those of E,* u ET = Eo u ET are mutually incident, there are maximal solvable Lie subgroups B, B* of G in flEEEOVE, c and flcoEEoE I o ~ * .Both Band B* are in a. Without changing a and H, by conjugation in a one may again arrange things so that H c B = B*. Thus the elements of Eo u El u E : are mutually incident. Since q~ relates elements of the same sort to one another, and q E , = E:, it follows that El = ET

After these preliminary transformations it will appear that Ei = E:

72.7.1

for all i ; hence b = b*, which is the assertion to be proved. Instead of 72.7.1, one may prove D, = D:

72.7.2

where D, = wEi, D: = WE:. Now suppose that this is true for all i cj with somej > 1. Then the line intervals [wa,wb] and [wa,wb*] meet the cell C,-l = CJ*-lbelonging to D,-, = D;-l in interior points The points of and u [t;-l,f;] are not separated by a hyperplane aL because such a hyperplane would have to contain one of the points hence both, hence either it contains to,which contradicts its separation property, or it is pierced by both line intervals. Thus this set is contained in some closed chamber that consequently contains D, as well as 0;.Since any closed chamber contains exactly one w-image of every sort, it follows that they are equal, which proves 72.7.2 for i =j . 72.8 The preceding theory is a practical tool for. classifying the pairs of elements of up to Int C. This is shown by a rather involved example in Section 73.

r

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN F4-GEOMETRY

439

73. THE CLASSES OF PAIRS OF ELEMENTS IN AN F4-G EOM ETRY

G E F4. Ordered second dressing supposed on H. Instead of a natural basis, 4 in H; to make things look an orthogonal one w I , ~ 2 , ~ 3 ,is~ chosen more symmetric, that is, such that

W*:

f w i ( i = 1,2,3,4),

P I = 3(Ul

p2 =

f W , ( i # j ; 1,2,3,4),

f w3 f w4).

w2

3(*Wl

W++:

*wi

- w2 - w3 - U413

- w3,

PI = 0 4 9 p4 = w3 - w4.

Fundamental weights : r l = w1, 772 = w 1 + w2,

7 3 =j ( 3 W l

+ + w3 -k W2

w4),

r4=2wl +wZ+w3.

On the basis

rwI,w2,w3,w41

theelements of n a r e

1 7 ( p l ) : rl,O,O,O1,...;

* r l , l , l , l l ,...;

17(p2): r1,1,0,01,...; 17(p3):

+ ‘3,1,1,1’,

17(p4): ‘2,1,1,01,

24elements. 24 elements.

. ..; . ..;

‘I, I , 1,01, .. .;

96 elements. 96 elements.

The dots indicate that all permutations of the coordinates and sign changes in the coordinates have to be applied. Ddescribes the geometry of the so-called 24-ceI1, a regular solid of 4-space. 17(p2), 17(p4), D(p3), 17(pl), respectively, gives the vertices, midpoints of edges, midpoints of 2-faces, and midpoints of 3-faces of this solid up to scalar factors. By geometric reasons the elements of T(P2)?

Q4),

T(P3h

are called points, lines, planes, symplecta, for which the letters a,

6,

C,

d,

440

68-75.

TITS GEOMETRIES

respectively, are reserved.

d c b a

.-.=.-.

PI

3

P2

P3 P4

The equivalence classes of pairs A, p E I7 with the corresponding 17(X,p)and U ( h ,p) = U;,oD1are exhibited. Incidence is indicated by a dash. There is a hierarchy of such incidence figures. If CP and CP’ are incidence figures for the same kind of pairs, the CP’ is called at least as weak as CP if the existence of CP‘ can be derived from that of CP by the axiomatic rules of 71.6.4. In 73.1-4 the classes of ‘X,pl are arranged from the weakest to the strongest figure. A subfigure of17(X,p) which is as strong as 17(h,p) itself is said to determine 17(X,p).Minimal determining subfigures are indicated in most cases. To simplify the figures “trivial” incidence consequences are sometimes left out. (A point and a plane are “trivially” incident if there is a line incident with both; a line and a symplecton are “trivially” incident if there is a plane incident with both.) Note the duality caused by the exchange of rays through 7r1 and n2,r3and n4.This reduces the task set. 73.1

x E r(p2),

73.1.1

h = -1 100,

el =+A + +p

(Point-symplecton.)

y E r(pl).

p = 1000.

= f.0100,

e2

=*A

+ Sp = 3

a

1100.

-1 100-0100-1 100-1000.

a-d-a-d.

h = 01 10,

73.1.2 =

p

fh + f p = 3 1110, 01 10-1

=

1000.

e2 =*A

++p

110-21 10-1000.

a-c-b-d. 73.1.3

h=1100,

p=lOOa

1 100- 1000.

a-d.

=

3 -2110.

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN

73.2

x

E

y E F(p3).

r(p2),

73.2.1

h = -1-100,

+ 3p = 3 & = +A + $1. = 3

( I = +A

-1-100--1-1

-1-1

*

0010,

(Point-plane.) p = 1 1 10.

10,

*

441

F,-GEOMETRY

(2 =

+A

+ +p = +

*

-1-120,

t4=+A + +p = 3 1120. *

10--1-120--0010-112&-1110

a-c-b-d-b-c. 73.2.2 51 =

+A

p = 1110.

X=-lOol,

+ +p = 3

*

(-3333 + -34+$),

f 3= fh + +p = 3 -011 1 ,

+ *P = + (01 12 + 2 -3+4+), 5 4 = +A + +P = * (01 10 + 2 - +if+).

52 =

*

+

See Figs. 8 and 9. 1111

FIG.8.

FIG.9.

The figure is determined by a-d-a-c. be uniquely extended to

Indeed, by 71.6.4(I), d-a-c

ATd\

d-a-c

can

68-75. TITS GEOMETRIES

442 and a-d-cl

to

cz-bz

/\/\ a-d-c,

;

the hidence bz-d is a trivial consequence.

73.2.3 f 1= +A

+ +p

=

+

A = 1-100, p = 1110. f z =+A +p = * (lo00 2-110), f 3 = +A + +p = 3 (1010 + 21 10).

+

+

-

+

*

(1010 + lOOo),

See Figs. 10 and 1 1.

FIQ.10.

\ ...... FIG. 11.

c

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN

F

4

-

~ 443 ~

The figure is determined by

a-d-b-c as well as by

a-b-a-c, since the lacking elements can be added by virtue of 71.6.4(1) or (4).

A = 1001,

73.2.4

5, =

+ 3p = 3 *(3333+ W ) ,

p = 1110. 52

=+A

See Figs. 12 and 13.

FIG.12.

FIG.13.

The figure is determined by

a-d-c.

+ 3p = 4

*

(21 10 + 2 *3M).

~

444

68-75.

73.2.5

A = 1100,

TITS GEOMETRIES

p = 1110.

1100-1110. u-c.

73.3

x

E

y

r(p2),

(Point-line.)

E T(p4).

A = -1-100,

73.3.1

[, = $A + $p = t & = +A

*

p = 21 10.

+ +p = 3 t4= fh + $p= $

f 2 = $A

-1-2-10,

+ 3p = 4 - 1-120,

*

0-1 10,

*

1010.

-1-2-10--0-110-1-120-1010-2110.

-1-100-

a-b-a-b-a-b. h = 0-1-10,

73.3.2 f I -- *A

+ $p = f

*

p = 21 10.

1-1-10,

52 =

+ +p

(]= -)A

=

+A

+ +p = 3 - 2-1-

1000.

0- 1-10-1 -1-1 0-2-1-10-1000-2

1 10.

a-c-b-d-b. h=0-101,

73.3.3

p=2110.

4, = ~ h + $ p = t . ( l - 1 0 2 + 2 . + - 3 3 f ) , (2

=+A

[]=+A

+ +p = +

*

(1001

+ 2 .-)-ff+),

+ f p -3 *2011,

[4=3h+~p=3'(1010+2.~3+f). See Figs. 14 and 15.

FIG.14.

10,

73. THE

CLASSES OF PAIRS OF ELEMENTS IN AN

FIG.15.

The figure is determined by

a-d-a-b, as well as by

a--6-a-c-b.

I n the first case d-a-b is filled up to

d-a-b

\A/ b-C

by means of 71.6.4(3) and then

a-d--6 to

a-d--6

//\/ b-a

by means of 71.6.4( 1). In the second case -6-a-c

is filled up to

and then, according to 71.6.1,

b-c--6 to

F

4

-

~445 ~

~

68-75.

446

x = 0-1

73.3.4

6, = ;3h + gp = 4

*

TITS GEOMETRIES

10,

p = 21 10.

6* = +A + 4p = 1010.

1-120,

0-1 10-1-120-1010-21

10.

a-b-a-b.

x = 1-100,

73.3.5

6, = gx + +p = +

*

(2 * lo00 + 2-1 lo),

p = 21 10. 62 = +h

See Figs. 16 and 17.

.

\

2-1 10

FIG. 16.

FIG.17.

+ 3p = f

*

(2 * lo00

+ 1010).

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN

F.,-GEOMETRY

The figure is determined by a-d-b.

x = 0110,

73.3.6

p

5, = * A + + =

= 2110.

1110.

0110-1 110-21 10.

a-c-b.

A = 1100,

73.3.7

p=2110.

1100-21 10.

a-b. 73.4 73.4.1

x

E

y E Q2).

r(pz),

(Point-point.) p

h = 1100,

= -1-100.

Rank 0. a, a.

x = 1100,

73.4.2

= j-x

= -1010.

f*=*x+fp=3.0110.

= jx++p=+*1210,

[,

p

+ j p = j- --1120. 10- -1 120- -1010.

1100-1210-01

a-b-a-b-a. 73.4.3

A = 1100, f,

= +A

p

=

1-100.

+ +p = 1000.

1loo-looo-1-100.

a-d-a. 73.4.4

h = 1100,

8, = $4

+ $1.

p = 1010. = 3.21 10.

1100-21 10--1010.

a-b--a.

447

448

68-75. TITS GEOMETRIES

73.4.5

A = 1100,

p = 1100.

1100-1 100.

a-a. 73.5

x

x = -2-1-10,

73.5.1

(Line-plane.)

y E r(p3).

E I'(p4),

p = 1110.

4, = $4

+ +p = -1000,

f2

=$A

f 3 =+A

+ += + .-i

f4

=+A

-2-1-10-

-1ooo-

1 lo,

-21 lo-

+ +p = + .-2110,

+ 3p = + .OllO.

-11 1 0 - 0 1 10-1 110.

b-d-b-c-a-c.

x = -2-101,

73.5.2

p = 1110.

+ +p = + - (-1001 + 4 -!& + -+ -#), =+A ++p = +-(-loo1 + 2 .++tf. + -2011), f 3 +A + -&= 3 - -101 1, & =+A + $p = + .(0011 + 2 *++# + -1021),

4, = +A 42

=

45 =

3x + +p = 4 *(0011+ t333 + -3+$4),

&, = +A + +p = + 0121, f7 =

f h + & = t * (01 10 + 3333 + 43+!z),

& = j-x +qp = j- .(0110 + 2 *-)&

+ 1120).

See Figs. 18 and 19. This is the first case in which IT@,,) is larger than nl(A,p),namely, by eight elements. This occasion will be used to build up l7(A,p) formally, starting with a chain which, by virtue of the construction, proves to determine the whole figure. From Fig. 18 one draws the chain -2-101-

-

p&- -2011-

-1011-

-1021-

4+3+4121-4+$+ -1 120-1 110,

for short,

b-~-b-~--b-~-b-~-b-~.

73. THE CLASSES OF PAIRS OF ELEMENTS IN AN

I

i

F

4

d

ri"

-

~449 ~ ~

450

68-75. TITS GEOMETRIES

73. THE CLASSES OF PAIRS OF ELEMENTS I N AN

F

4

-

~451 ~

~

452

68-75.

TITS GEOMETRIES

This chain is filled up by means of the algorithm in Section 71. (See Fig. 20.) One extends

b--b

/y'\b

to

( i = 1,2,3,4),

b-c-

c-b-c c-b-c

\1/

to

( i = 1,2,3,4),

dl

and after adding some trivial consequences

b,

/I \

a-c-a

to

d-b-d

to

a-c-a

( i = 1,2),

d-b--d

\I/

( i = 1,2); c1

finally, by virtue of 71.6.4(5), again after having added some trivial consequences, one adds to

whose elements are incident with both a3 and d2. the chain bl-c3--b3-c, All other incidences are trivial consequences and have been omitted in Fig. 20. The figure is determined by

b-c-b-c-b-c-b-c-b-c, but also by

b-d-a-d-C, and by

6-a-d-a-c.

h = -21-10,

73.5.3

p = 1 1 10,

c1 =+A + 3p = 3 ' ( 2 *-1100 + -11-lo), +A + +p = 4 (3 -1 100 + -12-lo), t3=-)A + = 3. (-1 100 + OlOO), =+A + +p = 3 * ( 3*0100+ -1210), f2 =

'

*

-)JL

(4

f5

=+A

&=+A See Figs. 21 and 22.

++/A

=

3 '(2 *0100+0110),

+ +p = 3 .(0110+ 2.1210).

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN

F

FIG.21.

FIG.22.

The figure is determined by b-a-d-b-c.

x = -1-102,

73.5.3'

5, =+A + 3p = 3

*

p

(0001

=

11 10.

+ 2 .-+ -3++),

~ 2 = ~ h + ~ p = ~ * (+OOll), 0001

t3= +A +

$/.L

=

+ +p

=

+

#+

+ 3 001 l), 3 . (2 -3333+ 001 I), *

(2 *

*

(3 * 3333 + 33+4>1

*

54 =

+A

65 =

ax + sp = a

56 =

f h +p = 4 . (4 . 3333 + 1 120).

See Figs. 23 and 24.

+

4

-

~453 ~

~

68-75.

454

TITS GEOMETRIES

-1

FIG.23.

FIG.24.

The figure is determined by b-c-a-d-c. It is the dual of 73.5.3. 73.5.4

A=-2101,

p = 1110.

+ +p = + *(-1101 + -1100 + 2 *-++++>, f2 = +A + 31. = + (-1201 + -1 100 + 4 -++++),

6, = + A

*

63 =

*

9+ 31.= 3 (-3333 + -+3+3),

t4= + A

*

+ +p

=

3

*

(0110

+ 021 I),

5 s = + A + QP = a '(2 *0110 + 3333 + +$+f), (a = + A + 4p = 4 '(2.01 10 + 2 *3+++ + 1210). See Figs. 25 and 26.

73. THE CLASSES OF PAIRS OF ELEMENTS IN AN

F

4

-

~ 455 ~

~

456

68-75. TITS GEOMETRIES

U

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN F4-GEOMETRY

457

Again n ( h , p ) is larger than f l ( A , p ) , namely, by four elements. To find a chain that determines the total number one can start with -2101-

-f-f-++-0110-1110.

It is denoted by b-d--a-c. From b-d--a and d-a--c one obtains b-a,-b,-a and d-cl-dl-c, respectively, together with the obvious incidences, and from b-d-c, and b,-a-c likewise c2,u2 and dZ,b2,respectively (see Fig. 27). After adding some trivial consequences one completes al-c2--a2, dl-b2-d2, and a-c1-a2 with elements b,, c,, and b4, respectively. Now one notes that d is incident with -a2

-a,--b,--a

\

which implies the existence of

by 71.6.4( 1); from this b3-c4 and b4-c4 follow easily. Similarly, if c' is defined by d-bl-dz

\I/

9

C'

one finds b' with c1--a-d,

\ I / b'

9

whence by reduction with respect to d, apart from

b,--a-c

\ /\/ c4-b4

6,-a-c, and

\/\,/ , c'-

b

whence b'-c,, b'-c,; therefore c' = c4, 6' = b4, by 71.6.4(5). This shows the crucial incidences d2-c4, b4-c,. The remaining incidences are trivial consequences.

458

68-75. TITS GEOMETRIES

V

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN F4-GEOMETRY

The figure is determined by

b-d-a-c.

x = 1-102,

73.5.5

p = 11 10.

5, = 3h + +p = 3 '(2 1001 + t -+3+ +++-I$>, *

~ z = + x + + p = + - ( 3 * 1 0 0+12 . + - & + 1012), 5, = +A + +p = (1001 + 101l), t4=+A +p = 3 '(2 -+J& 201 l), 4 s = *A $p = * (3 * +ti+ -I- 1010 ++++), & = f-h+ $1.= '(4 *++# + 1010 + 2110).

+ +

+-

*+

+

+

See Figs. 28 and 29. 1 1 1 1

FIG.28.

FIG.29.

459

460

68-75.

TITS GEOMETRIES

Building up from I-l02--1001-;~~~--I

110,

(see Fig. 28), or, for short, b--n--cl-c,

one gets, in Fig. 30, consecutively, the following elements, with incidences as shown:

FIG.30.

by 71.6.4(3) h--cl--bI--rl the reduction result of (I, by 7 1.6.4(5)

in the reduction result of a and a-cz-bz-c

in

in the reduction result of d, and furthermore dl and a l by means of 71.6.4(6). The incidences not shown in Fig. 30 are trivial consequences. The figure is determined by b-a-d-c .

A=-2110,

73.5.6

el

=

+A + +p = - 1

110,

p = 1110.

[*

=

*A + ; p

-21 10- -1 I10-0110--1110.

6-c-a-c.

= 01 10.

73.

461

THE CLASSES OF PAIRS OF ELEMENTS IN AN F4-GEOMETRY

h = 2-1-10, p = 1 1 10. 5 2 = *A 4p = 3 * 21 10. 4, = +A + +p = 3 * 1oO0, 2-1-10-1000-21 1 0 - 1 110. b-d-b-c. This is the dual of 73.5.6.

73.5.6 '

73.5.7

+

x=2-101, p = 1110. 5, = +A + +p = 3 * (1000 1001 2 -3-+++), t2= +A + +p = (1000 + 201 l), 5 3 = +A +p = 3 * (1010 2 *$+++), ~4=~h+~p=*'(1010+2*+++++2~2110).

+

+

+-

+

+

See Figs. 31 and 32. 21 10

10

2-

\\\

/

0

3 111 0 ----2 222

'

FIG.31.

FIG. 32.

462

68-75.

TITS GEOMETRIES

The figure is determined by b-c-b-c--b-C. 73.5.8

A = 1102, p = 1110. 5, = 3h 31. = 3(1101 + 2 333t ),

+ + +p = 3

52 =

*

(1 100

+4

*

++++)

See Figs. 33 and 34.

\

1101

\

.

1100

FIG.33.

a

FIG.34.

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN

F

4

-

~463 ~

The figure is determined by

b-d-C. 73.5.8'

x=21-10,

p = 1110.

4, = j x + jp = j * (2 lo00 + 1 loo), t2= +A + +p = + . (21 10 + 2 1loo). *

See Figs. 35 and 36.

FIG.35.

..........

I

b'

FIG. 36.

c

~

464

68-75. TITS GEOMETRIES

The figure is determined by b-a-c, This is the dual of 73.5.8.

73.5.9

x=2101,

p = 1110.

(, =+h++p=+*(333)+ llOo+;fi+)? t2=+A + +p 3 . ( 2 .a++ + 1100 + 21 10). =

See Figs. 37 and 38.

Fro. 37.

FIG.38.

The figure is determined by

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN F4-GEOMETRY

465

as well as by

b-C-b-C. 73.5.10

x=2110, p = 1110. 21 10-1 110. b-c.

73.5.1-10 From the weakest to the strongest relation on pairs rb,cl the hierarchy runs:

73.6

x E r(p4),

y

E r(p4).

x = 21 10,

73.6.1

p

=

(Line-line.)

-2-1-10.

Rank 0

b, 6. 73.6.2

h = 21 10,

p

=

-2-101.

5, = +A + 3p = 3 - (2 3333 + lolo), + 3p = 4 (2 ’ 3333 + 1021), *

5 2 = 3h

*

t3= +A + 3p = 3 -0011, 54 =

+A

+ 3p

=

+

t5= +A + +p = 3

* *

(2 * -3 -33-5 (2 -3 -4& a

+ -1012),

+ -1001).

See Figs. 39 and 40. -1001

21

.

1010

.

’ FIG.39.

68-75. TITS GEOMETRIES

466

FIG.40. L!(A,p) is four elements larger than f l ( A , p ) . a2

.

.

a1

FIG.41.

Figure 41 shows that the figure is determined by b-d-a-d-b.

73.6.3

A=2110,

p=-2-110.

5 , = $A + )p = f (1010 + 111% E* = +A + +p = + - (1010+ 1 120), 5, = +A + +p = 0010, *

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN

0110

FIG.42.

F.,-GEOMETRY

467

468

68-75.

TITS GEOMETRIES

17(h,p) is four elements larger than nl(h,p).

/ b3

/ 4

/

’\

\

/ /’ /’

FIG.44.

Figure 44 shows that the figure is determined by

b-a-d-a-b. As a matter of fact, this is equivalent to

b-a-c-b-c-a-b because of

b-a-c-d-c-a-b, b-a-d-a-b, b-a-c-d-a-b,b-a-c-b-c-a-b Note, however, that this chain cannot be realized in I‘O.

73.

THE CLASSES OF PAIRS OF ELEMENTS IN AN

73.6.4

h=2110,

5,=$A

+*p

=

*

*

F

4

p=--2101.

(2

.++++ + 2

1110

+ lloo),

+ +p = + (2 ++g + 1210), 5 3 = qx + 31. + (2 at$++ + 2 +f++ + OllO), E*

= $4

*

=

*

*

*

t4= fh + +p = f ,0211, tS= +A + * p = (2 -p& 66 =

+A

+ 3p =

5, = f h + $p =

+ + 2 .-+f++ + OlOl), + (2 ++++ + -1201),

*

' *

*

(2 *++

See Figs. 45 and 46.

FIG.45.

FIG.46

+2

*

-1101

+ -1100).

-

~ 469 ~

~

470

68-75. TITS GEOMETRIES

nl(A,p) falls short of n(A,p) by four elements. The total figure is also determined by

b-C-b-C--b-C-b-C-b as well as by

6-d-C-d-b. From the latter figure, the complete figure is reconstructed in Fig. 47, in which the chains b-cl-al-c, c-a2-c2-b, al-bl-a2, cl-b2-c,-bl, and b--a3-b2, b3-a4-b, c2-dl-c3, and finally a3-b,-a2, b,-c3-b3-c2 al-b5-a4 are retrieved consecutively with the incidences shown. The remaining incidences are trivial consequences.

FIG.41.

73.6.5

x = 21 10, p = -1-2 [I =+A + +p = 1010,

10.

*

+ +p = 1-120, 6 3 = +A + +p = 0-1 10.

52

=+A

21 10-1010-1-120-0-1

1 0 - -1-210.

b-a-b-a-b. 73.6.6

x = 21 10, p = -12-10. el = $A + i p = (3 1100 + 2 1 1 lo), t2=+A + +p = 3 '(2 1100 + 1210), *

* *

*

471

73. THE CLASSES OF PAIRS OF ELEMENTS IN AN F4-GEOMETRY f 3=+A +*p

=

f * (1100 + 2.0100),

t4= +A + 3p = 4

*

(12-10

+ 6 . OlOO),

t5=+A + +p = + *(01-10 + 4 *OlOO). See Figs. 48 and 49.

\

/

. . :. . .: . . l

\ 2 1 1 0 \/ l

\

12-10 \ : - 1O'

-12-10 -1 0

/ /0100

1

I /

1 1 0 /,,lo

FIG.48.

a

\

......

C

/ FIG.49.

The figure is determined by b-a-d-b.

68-75.

472

TITS GEOMETRIES

A = 21 10,

73.6.6’

+ +p = +

p

=

1-2-10.

+ lolo), & = +A + +p = * ( 6 lo00 + 2-1 10)s f 3 = + A + +p = 3 (2 1000 + 1-loo), t4= +A + +p = + (2-1-10 + 2 - 1-loo), t5= ) A + $p = 4 -(2.1-1-10 + 3.1-100).

(, = +A

*

(4.1000 *

*

*

FIG.50.

b

\

/ .......

\

/b

FIG.51.

7 3 . THE CLASSES OF PAIRS OF ELEMENTS IN AN F4-GEOMETRY

473

The figure is determined by b-d-a-b. 73.6.7

5, = j h + &.L

h = 21 10, =

1110,

52 =

p = -21 10.

fh + f p =0110,

63 =fh

21 10-1 110-41 lo- -1 1 lo-- -21 10.

b-c-a-c-b. h = 21 10,

73.6.8

[,

= $A

52

=+A

p

=

-1201.

+ * p = 3 - (p& + 1100 + 1 1 lo), + +p + (2 .+f$f + 1100 + 1210), =

*

tx + 31. = ft3f. t4= +A + +p = 3 . (2 0100 + 0101 + 021 l), 53 =

*

tS= ah + $p = +.(0100 + 0101 + -+&>. See Figs. 52 and 5 3 .

+ * p = -1110.

68-75.

474

TITS GEOMETRIES

F I ~53. .

The figure is determined by b-C-b--C-b-C-b, as well as by the conjunction of

b-d-a-b

and

p = -1210.

A = 21 10,

73.6.9

(,

= $A

f2 =

+A

b-a-d-b.

+ i p = 4 '(4.1110 + 1100),

+ +p = 3

*

(2 11 10

+ 12101,

g3 = +A + +p = + ' (01 10 + 1210), & =+A + +p = + * (3.0110 + 2 *OlOO>. See Figs. 54 and 55.

73 THE CLASSES OF PAIRS OF ELEMENTS IN AN F

21

FIG.55.

The figure is determined by b-c-a-b.

h = 21 10,

73.6.9’

p = 1-210.

+

,$, = +A + +p = 3 * (3 * 1010 2 * lOOO), t2= +A + +p = * (1010 2-1 lo), = + A ++p = 3*(2.1-110 2-1 lo),

c3

f4 =

See Figs. 56 and 57.

+

+

$A + $p = 4 . (4 * 1-1

+ 10 + 1-100).

4

-

~ 475 ~

~

476

68-75.

TITS GEOMETRIES

FIG.5 1

The figure is determined by

b-a--c--b 73.6.10

x=2110,

5, =

p=2-1--10

+ 31. = 2

1000. 21 10-1000-2-1-10.

b-d-b.

*

73. 73.6.11

THE CLASSES OF PAIRS OF ELEMENTS IN AN

F

4

-

h = 21 10, p = 2-101. = j h + +p = 3 - (2 * 1000 + 10 10 + 2 * +#), [* = +A + f p = 3 * (2 * 1000 + 201 l), &=+A+ jp=33(2’1000+1001+2*~t3~).

[,

See Figs. 58 and 59.

21 10

1010

-2011 -1001 FIG.58.

FIG. 59.

This is a special pair of lines in a symplecton. The figure is determined by

A\

h-c-a-b as well as by

9

~ 477 ~

~

478

68-75. TITS GEOMETRlES

73.6.12

A-2110,

5, = +A

p = 12-10

+ +p = 3

*

(3 - 1 100)

2110-1 100-12-10.

b-a--6.

x

73.6.13

p = 21-10.

= 21 10,

~,=-)x+~p=1000+1100.

See Figs. 60 and 61. d

1000

FIG.60.

73.6.14

FIG.61.

x=2110,

5 , = +A

p=2101.

+ 3p = 3

*

(1 100 + 2 * %-H).

See Figs. 62 and 63.

FIG.62.

FIG.63.

74. THE INCIDENCE GEOMETRIES OF REAL GROUPS

73.6.15

h=2110,

479

p=2110.

2110-2110.

b-b. 73.6.1-1 5 From the weakest to the strongest relation on pairs of lines the hierarchy runs 7 10 / \ / \ 2-4-6-8-9-11

'3-56'

9'-12

74. THE INCIDENCE GEOMETRIES OF REAL SEMISIMPLE LIE GROUPS

The theory developed in Sections 70 and 71 for complex semisimple Lie groups can be adapted to the real case. G E Gru Lie Lin Com SS, G its infinitesimal algebra, C an involutory semimorphism, and Gc, Gc the C-restrictions; other notations as usual. Gd sub Gc is the C-restriction of C-invariant G' sub G. 74.1. Definition Gd sub G, is called parabolic if C' sub G is parabolic. The same for GL. Note that parabolic GL need not comprise a maximal connected solvable subgroup of Gc, though G' does so with respect to G. Proposition If GL is maximal parabolic sub G,, then G' = M n CM for some maximal parabolic M sub G. Proof For Mtake a maximal parabolic subalgebra of Garound C'.Then CM is of the same kind, and M n CMis C-invariant and thus equals G'.

74.2. Theorem If Gd and Gi are parabolic sub G,, then G' n G" contains a minimally compact C-invariant trunk of G. Proof Let B',B" be maximal solvable sub C',G", respectively. Bruhat's lemma provides a trunk contained in B' n B" c G' n G", thus a G-regular element /to of G' n G". Then 7h0 + X h o is still regular for suitable T and lying in G' n C " .Moreover, it is C-invariant. Therefore it defines a C-invariant trunk H of G within G' n G".

480

68-75.

TITS GEOMETRIES

Take C,, on the same trunk after C-third dressing, that is, with CC,, = C,,C (see 51.6). Suppose H is not yet minimally compact. Then for some a = CC,,a the condition of 60.9 is not satisfied, Therefore there is some a = CC,,u

E

W*

with CCunea= -e,;

thus Ce,

= e-,.

e, E G', for otherwise both e, and e-, 4. G', which contradicts parabolicity. The same is true of G". Therefore e,,e-, E G' n G". This shows that U,, as used in 60.15 to transform the trunk H, can be formed within Int(C' n C " ) . The procedure in 60.15 produces a C-invariant trunk of G within G' n G", though now with a larger multiplicity of the eigenvalue -1 of CC,, on the trunk. This proves the assertion. 74.3 In the sequel the trunk H is supposed to be C-minimally compact. By 60.17 it does not matter which one is chosen. C,, is taken with respect to H , and with CC,, = C,, C. Under the aspect of reality the incidence geometry of Gc should be defined as consisting of the maximal parabolic subgroups of Gc. An equivalent and more convenient approach is to use their complex extensions. This explains the following:

rc

Definition is the set of C-invariant Lie subgroups of G whose C-restricis endowed with the following tions are maximal parabolic sub Gc. The set incidence relation : a,b are incident if a n b contains a maximal connected solvable subgroup of G. the incidence geometry of Gc, is considered as acted on by Gc through its adjoint action. Its transitivity classes are the sorts of the geometry. is the set of elements of rccontaining the (C-minimally compact) trunk H. Note that by 74.2a given pair of elements of may be supposed to be in thanks to the action of Gc on

rc

rc,

rO ,

rc

r,.

r,O,

74.4 To continue as in 70.6 one must redefine wand l7. By 74.1 any a E F: can be obtained as x n CX with x E To.By 70.6.2 the infinitesimal algebra of x is spanned by H a n d the ear,with (wx, a ) > 0. Clearly wCx = C* wx..Thus the infinitesimal algebra of a is spanned by H a n d the e, with 74.4.1

( w x , a) 2 0

and

(C* wx, a) > 0.

74. THE INCIDENCE GEOMETRIES OF

Now x and Cx are incident as elements of parabolic. Therefore by 70.6.1 (wx, a)(C* wx, a ) 2 0

REAL GROUPS

48 1

r, for their intersection is still

for all a

E

W*.

In a more compact way the condition 74.4.I can now be written a) 2 0 (wx + C* IYX,

or, equivalently, ((1 - CC,,,) WX,a ) 2 0.

This suggests the following: Definition 17, is the set of (1 + C*)X with X E I7 (i.e., in the set of equivalents of fundamental weights) such that X is incident with C*X. wc is defined by w,a = (1 - CC,,)wx

O , with a = x n Cxand x E To. for a E r The last part of this definition is still to be justified. A little more will be proved :

ro,

Proposition If x,y E x incident with Cx, y incident with Cy, and x n C x c y n Cy,thenx=yorx=Cy.

Proof By 70.10, since y 1 x n Cx, all parts containing wx and wCx also contain wy. The smallest part, that is, the cell, containing wx and wCx is oneor two-dimensional and its only members of I7 are wx and wCx. Therefore wy = wx or = WCX,which proves the assertion.

74.5 The following was proved in 74.4: Proposition wc maps F,O one-to-one onto algebra of a is spanned by Hand thee,,, with

nCsuch that the infinitesimal

(Wca, y ) 2 0.

Indeed w C maps onto 17,: Given some X E 17 incident with C* A, there is an x E T oincident with Cx and such that X = wx. Now x n Cx is contained in the complex extension y n Cy of some maximal parabolic subgroup of Gc so that x n Cx = y n Cy E I',O with y E ro.But then by Proposition 74.4, x n Cx = y n Cy,which proves the assertion. 74.6. Proposition (1

+ CC,,)a

E

W* for no a

E

W*.

482

68-75. TITS GEOMETRIES

Proof CC,, multiplies [e,, CC,,e,] by -1 ; thus by 60.9 and the minimal

compactness of H this expression cannot be a branch and consequently (1 + CC,,) a cannot be a nonzero rootform. 74.7. Definition The intersections of parts and cells of H A with H,* are called C-parts and C-cells. A C-cell of maximal dimension (=real rank) is a closed C-chamber, its interior in Hg,he a C-chamber. A C-edge is a onedimensional C-cell. Jnt, W* is the maximal subgroup of Int W* leaving H z , h e invariant, and restricted to H z , h e . Theorem Int, W* is simply transitive on the set of C-chambers. Its elements are induced by inner automorphisms of C, preserving H. Proof Take two C-chambers with a common wall of codimension 1 in H:,he. The wall is determined by some a E W*. Then a # CC,, a,for otherwise a(H,-he) = (0). For the transitivity assertion it suffices to show that the reflection in this wall is in Int, W*. There are three cases with respect to a:

74.7.1 CC,,a = -a. Then with S, defined by 20.9, C*SaC* thus Sa itself induces an element S of Int, W* as required.

= S,,,

= S,;

74.7.2 (1 - CC,,)a E W*. Then C*S~I-ccu.)aC* = S(I-ccun)a. Since the orthoplanes of a and (1 - CC,,)a have the same intersection with f ? z , h e , an element Sof Int, W* as required is furnished by S(,-ccu,,)a. 74.7.3 CC,,a # -a, (a,CCun a) = 0,

(1 - CC,,)a

4 W*. Then, as implied by 74.6,

sa SCCYna= &c..a C(Sa SCCuna)C = &Cuna

sa,

sa = sa &c..a*

Since S, SCCuna a = -a, S, SCCuna induces an element S of Int, W* as required. To show that Int, W* is simply transitive, take an S E Int W*, leaving Hz,heand a closed C-chamber Kinvariant. Scauses a permutation in the set D of the closed chambers that contain K. The reflections in the common walls of elements of D produce a subgroup F of Int W*; the elements of F leave K, hence elementwise invariant, and Facts transitively on D. Thus there is an S‘ E F such that S-I S’ leaves a certain element of D invariant. Since Int W * is simply transitive on the set of all chambers, S - ’ S ’ = 1 ;thus S = S’, and S restricted to H z , h eis the identity. To prove the remainder of the theorem, one must extend S as obtained in

74.

THE INCIDENCE GEOMETRIES OF REAL GROUPS

483

74.7.1-3 to the whole of C according to 33.1, and then replace it by some S' = h S k ' such that h = exp h, h E H , and S' E Int G,:

+ 2-,)

74.7.1' S = exp .(Pa

S'

Ce,

because of CC,,

with a certain imaginary T.

=

Ash-' = exp(Th(2,

=

exp(.r(exp a(h) C, Ce-,

= Ke,,

= C,,

+ P-,) h-1)

+ exp(-ot(h))

P-,)).

with

KI?

= ie-,

=

1,

C. For S' to be real, one gets the condition on h:

-

exp(a(h) - a@)) = - K , which can easily be fulfilled. 74.7.2' The same argument applies to (1 - CC,, a) instead of a.

+

74.7.3' Neither (1 CC,,)a nor (1 - CC,,)a are rootforms, not even 0. Hence S' is the product of exp .r(exp a(h)l, and exp T(exp -a@) P-,

+ exp(C* a)(h)Pc.,) + exp(-C*

a)@)iLC.,),

7. If Ce, = ~ e ~ then . ~ ,Ce,., Ce-c.ar= K-' e-,. Furthermore, ~i= 1.

with a certain imaginary

= K-'

e,, and Ce-,

=

The reality required boils down to -

exp((C * a) (h) - a(h))= -K, which can be fulfilled, since a and C* a are linearly independent. 74.8 If A E I7 and C* A are incident, there is an at most two-dimensional cell that contains both. This cell intersects HZ,hein a C-cell which is one-dimensional, that is, a C-edge. (1 - CC,,)A lies on this C-edge. The converse is also true: Proposition The C-edges are the sets of nonnegative multiples of elements

of 17,. Proof Let d be a C-edge and D the smallest cell that contains d. The C-

invariance of d implies that of D . Choose A E 17 on D . Then C*h on D , h + C*h#O, and h + C*h on D as well as in HZhe. This shows that d is spanned by h + C*h = (1 - CC,,)h.

484

68-75. TITS GEOMETRIES

74.9. Proposition The following statements are equivalent: 1. a, b E I'g are incident. 2. There is a C-cell containing w,a and wcb. 3. ( w c a , y ) ( w c b , y )2 0 for all y E W*.

ro,

Proof If a = x n Cx, b = y n Cy with x, y E then incidence of a,b means that x n Cx n y n Cy contains a maximal solvable Lie subgroup, thus that x, C X , ~Cy , are mutually incident, and that their w-images are contained in a common cell; this, however, is equivalent to wca, wcb being contained in a common C-cell, This proves the equivalence of 1 and 2. The equivalence of 2

and 3 is obvious. Definition Ac,pc E 17,are called incident if

(A,, a)(pc,a) 2 0

for all

r,

tl

E

rO ,

W*.

74.10. Theorem Every sort of intersects and even has a member whose w,-image is in a given closed C-chamber. Two elements of belong to the same sort iff they are equivalent under a C-real element of Int(G,H). The number of sorts equals the real rank of Gc.

rO ,

r,

Proof The infinitesimal algebra of a E contains a C-minimally compact trunk H' of G (see 74.2). By 60.17 H' is carried into H by the action of Int C,.

The same action carries a into an element of I'$. This shows the truth of the first sentence. The second follows from theorem 74.7. If a, b E r O , are equivalent under Int C,, one must show that they are also equivalent under an element of Int Gc that leaves H invariant. By the foregoing w,a, w,b may be taken in the same closed C-chamber, that is, incident. Now a = x n Cx, b = y n Cy with x,y E x incident with Cx, y with Cy. By Proposition 74.4 and the G,-equivalence of a, b one may suppose that x,y are equivalent under Int G. Thus wx, wy are equivalent under Int W*. They are also incident, hence equal. This shows that a,b are equal. Finally, two distinct C-edges of a closed C-chamber cannot be Int, W*-equivalent, since otherwise two distinct edges of a closed chamber would be Int W*-equivalent. This shows that there are exactly as many sorts as there are C-edges of a closed C-chamber.

ro,

74.11 A C-chamber can be represented by a graph with dots corresponding to the walls and multiple bonds corresponding to the angles of the walls in the usual way. With due regard to the lengths of the related combinations of rootforms, one can even put arrows in the bonds. It will appear that no new graphs occur except those found in the complex case. Definition

By C-graph of Gc one means that of a C-chamber of G,.

75. C-GRAPHS OF INCIDENCE GEOMETRIES

485

74.12 After this successive reduction of incidence problems in

rcto those in

T': and ITc and finally to problems on the C-graph of Gc, one can repeat the analysis of Sections 71-73 with slight changes. The most notable deviation is that arbitrary trunks are restricted to the minimally compact ones. The incidence geometries belonging to real semisimple groups provide a new example of a Tits geometry. In particular, those of real rank 2 give new generalized m-gons (m= 2,3,4,6).

75. C-GRAPHS OF INCIDENCE GEOMETRIES OF SIMPLE SEMISIMPLE LIE GROUPS

When studying the incidence geometries of particular real semisimple Lie groups, one can restrict oneself to the simple case. The compact types can be dismissed because then C* = -1 and X E 17can never be incident with C*X = -A, which causes II,, hence to be void. Standard type geometries behave as the corresponding complex ones. In the twin case Gc E L,, there is a trivial isomorphism between r ( o f G E L) and r, (of Gc E L**). This justifies the supposition : Gc noncompact, nonstandard, nontwin, simple.

r,,

75.1 H is a minimally compact trunk with a real order (see 60.18). G is in C-third dressing on H . Thus,

for j = 1, . . ., k,

c*PJ = -PJ c*Pj = P j mod PI,

for j = k + l , ..., m,

. . .)pk

C*pj = P ~ + , , , + ~ mod - ~ P I , . . ., pr

for j

=m

+ 1, .. ., 1.

For the fundamental weights this yields C * r j = - r j mod

rk+l,

. . ., rl

for j = 1, . . ., k,

C*?Tj==77 J

for j = k + l , ..., m,

C*rj

for j = m + 1,

=ri+m+l-j

..., 1.

The nonnegative multiples of the rj+C*rj

with j = k + 1, ...,+(f + m )

are elements of the constituting edges of a C-chamber. Its C-graph is constructed in any particular case. At any dot a primitive rootform p is placed, together with C * p if they are different. Note that the corresponding wall is Pl

"

HC*.lle= ( C * P ) l

"

fchc.

The real rank I' of G, is t(f+ m) - k.

486

68-75.

TITS GEOMETRIES

75.2 To get more insight into the details by which incidence geometries with the same graph are distinguished from one another more data are added: Take I' mutually incident elements a,, ..., a,. of a, from the sort F,,,; for example, those mapped by wc into n k + j + C * ' r r k + , , j = 1, ..., I'. Then for p E { 1, ..., 1'}consider the manifold V, of all

r,,

x

E

rC,,,x incident with all a, ( j# p ) .

For j = 1, ..., I' let Gc,, be the normalizer in Gc of a,,, (a,,, is the group infinitesimally generated by the C-restriction of the infinitesimal algebra of a,); that is, Gc,, is the stabilizer of a, in Gc as acting on Then a,,, is the I-component of Gc,,. Let x E F,,,. Then both a, n n,#,a, andx n nJspa,are parabolic because of Proposition 70.7; hence by Theorem 74.2 their intersection contains a minimally compact C-invariant trunk H' of G. By the G,-conjugacy of those trunks and by Theorems 74.10 and 74.7 there is an element g E G, such that

r,.

g'H'=H,

for j = 1 ,

&,=a,

..., 1';

+

therefore g E fly=l G,,,. Now wcgx is incident with all w,a, = T k + j C*nk+, with j f p , thus either coincides with w,ap = 'rrk+p c * ' r r k + p , from which follows, gx = a, and x = a,, or is the image of nk+, C * n k + p under the reflection in the wall in HZ,hc determined by P f + p ; now the element of Gc constructed in the proof of Theorem 74.7 effectuating this reflection is easily seen to lie in the 1-component of n,#pa,,c. Hence V, is transitively acted on by n,+,GC,,, even by its 1-component. Therefore V, is connected and, as acted on by Gc, isomorphic with the homogeneous space

+

n

i#P

+

G c , , / ~Gc.j . I

This quotient can be reduced by the maximal normal Lie subgroup of the numerator contained in the denominator. In the complex case this is the radical of the numerator. After this reduction a semisimple group of rank 1 is left in the numerator with a maximal solvable Lie subgroup in the denominator. This means that all V, in the complex case are essentially complex projective lines. In the real case, after factoring out the radical of the numerator, the numerator and the denominator may still contain a nontrivial common direct factor, part of the isomorphic image of the C-restriction of the group infinitesimally generated by the e k p , ,.. ., ekpr.This common factor is factored out also. Since V, is connected, all that finally matters is the I-component of the numerator, denoted by G,*,,. It is indicated in all particular cases. It is, of course, of real rank 1. V p is a homogeneous space of C;,, with respect to a maximal solvable Lie subgroup.

75.

C-GRAPHS OF INCIDENCE GEOMETRIES

487

75.3 The real groups of real rank 1 occurring in this connection and the related manifolds are (cf. also 75.19) :

complex projective line (or quadric in real projective 3-space). quaternion-projective line (or quadric in real projective 5-space) real projective line. hermitean quadric in complex projective k-space. quadric in real projective 2k-space. hermitean quadric in quaternion-projective (k - 1)-space. quadric in real projective (2k - 1)-space. octavian projective line. (The last statement will not be explained.) All quadrics are of the type tc- * .+ -. 75.4 Note that because of the choice of a real order the basis of the (-1)eigenspace of CC,, will be a bit different from that in 60.23. The deviations are : Al,j,

l ' = j < + ( l + 1):

~ , - w , + ~ - , ( i 1, =

Ai,o,*, l o d d , I ' = f ( l - l ) : BiJY

1 ' = 2 ( j - 1) =21-2j+3 =1 w , ( i = 1, . . ., 1').

Cl,j ,

I'

Di.,,

3 <j < tl+ 2, I'

fl:

=j
3 ( / + 1 ) for j = I :

w21-l

1' = [fl]: w Z i -I Dl,l, instead of w l - ] w I .

..., 1'),

488

68-75. TITS GEOMETRIES

By use of the kaleidoscope group it is easy to derive from 60.23 that C* can be assumed to act as above (of course, since C* is involutory and orthogonal on H&,it is now uniquely determined). In 75.5-18 it will be seen that the established order is indeed real with respect to C*, though the old enumeration of W + +is used. 75.5

I'

Al,,,

= j < +(l+ I).

C * maps :

..., 1'; 1 + 2 - l', ..., 1 + I , i = I' + 1, ..., 1 + 1 - 1'; i = 1,

w , += - w ~ + ~ - ~for

for

0 , -+ - 0 1

or:

for i = 1, ..., 1 ' - I ;

pi + P , + ~ - ~ PI,

Pi Pi+l-I'

+

PI'+I

-+

-pi

+

Pi,

1+2-1',

..., I,

+ Pr+2 + * ' + P l + l - l , , for i = I' + 1, ..., 1 - l', *

+ Pr+1 + * + P I - I , . * *

. . ---A,**

-A1**--A1**-

* AI+2-21'.1.

C-edges : rI+

75.6 A2,-l .o,*,

I'

T ~ + ~ - ,

for i = I ,

..., 1'.

= m.

C * maps :

for i = 1,

021-l-w21

..., 1'+ I ,

or : for i = I ,

p2i-I + -p21-1 ~ 2 + 1 P~I-I

+ p2t + P

..., 1' + 1,

~

~ for + ~ i = 1,

.. ., 1'.

-

..._

P2v

C-graph : P2

PI

P4

+ P2 + P 3 P 3 + P4 + Ps

C-edges : 2r2,

P21'-l

for i = 1,. . ., I'

+ P2V + P 2 I + I

75. C-GRAPHS OF INCIDENCE GEOMETRIES 75.7 Bl,,, I ' = 1 f o r j = 1, 1 ' = 2 j - 2 j > + I + 1.

for 1 c j < + l + 1, l t = 2 1 - 2 j + 3

C * maps : w i -+ w f

for i = 1,

w f 3 -wf

for i = I'

. . ., I', + 1, . . ., I ;

or: for i = 2 , 3 ,..., I',

PI+P~ P l ' + l +-P

l

P

t

l

+ 2p,,+2 +

* * *

+ 2pi + 2p1,

for i = I ' + 2,

pi +-pi

. . ., I, 1.

C-graph :

.

Al,l-Al*l-

*

-

* Bl-l~+l,l.

+,I

C-edges :

for i = 2, . . ., I'

27rf

75.8 C l , j ,I'

=j

+ 1.

G $1.

C* maps: P2f-I

P2r PZI,

+

-+

. . ., I', i = 1, . . ., I' - 1,

for i = I,

-PZf-l

-+

+ pzt + pzfi.l for ~ 2 1 f - 1+ + 2 ~ 2 1 3 ~+1 . + 2p1-, + p1 2Pl-, + PI PZI-I

~

2

1

or : w 2 f - ,t

t

for 21' < I,

* *

5

w2f

w,--f-wi

for 21'= I ;

for i = 1, . . ., I', for i = 2 1 ' + 1,..., 1.

C-graph : p2 PI

+ P2 + P3

P3

P4

+ P4 + p s

-

... p21'-3

P21'-2

+ P21'-2

5

f P,?l'-l

489

for

68-75. TITS GEOMETRIES

490

for 21’

=I:

C-edges: 27r2,

75.9 DI,,, 3 Q j < 31 D1,o,*,I‘

=

for i = 1,

...,1’.

+ 2, I’ = 2j - 4, and

1 , and

DI,,,*,3<j P I . P2

where D3,0,*,DzSo,*are understood as A3.0,*, A,,**. C-edges: 27r1+2

for i = 1,

., ., I’

, with

7rI

+

7r2

for l’= I - 1

75. C-GRAPHS OF INCIDENCE GEOMETRIES

75.10 D,,,, Zodd, Z'=+(Z-

49 1

1).

C* maps w21-l

for i = 1, . .., I',

w21

-+

w1 + -cot;

or: P21+1

for i = 1,

+ ~ 2 t +p2t+1

for

~ z + t ~ 2 t - I

PI

--f

. , ., Z', i = 2, . . ., Z',

I

-P21t

+

P2 + P I ,

+ PI.

P2 - + P I

C-graph :

-

P4

P5

P3 f P4 f f'5

P6

+ P6 + P7

- ... -

A3.0,*-A3,0,*-

PI-1

Pt-2 * * *

-A3.0,*

C-edges : ( i = 2,

75.11 Dl, I , Z even, I'

=

. . ., Z'),

+ P t - l + PI * A3vl

PI P2 + P I

+ n2.

n1

+Z.

C* maps:

or: P21+1

+

+ +P + PI + P2,

~ 2 + t ~21-I PI

PI-l

-+

..., 1 ' - I , ~ i = 2, . . ., I' - 1,

for i = l ,

-P21+1

~

~ for +

PI -+PI,

P2

+

-P2.

C-graph:

C-edges : 2m1

for i = 2.

. . .. 1'. and

2n,.

68-75.

492 75.12 E6,1, 1'

TITS GEOMETRIES

= 2.

C* maps:

C-edges :

75.1 3

Eb.2,

I'

= 4.

C* maps:

C-graph :

C-edges :

75.14

E6,0,*,

I'

= 2.

C * maps :

PI -+PI f 3 f 2

+

-I329

+

P3

+ 2p4 + 2P6 + p2 + ps, + 2P5 + 2P6 + p2 + p4,

p4 + -p4,

p5 -+7 3 5 ,

p6

+ -p6.

75. C-GRAPHS OF INCIDENCE GEOMETRIES C-graph P3

Pl

PI

+ 2p4 + 2P6 + P2 + P S P 3 + 2p5 + 2p6 + p2 + p4' D5,0,*-D5,0,*.

C-edges : 2T,,

75.15

2773.

1'=4.

E7,1,

C* maps:

75.16

I'

E7,2,

= 3.

C*maps:

C-graph :

C-edges : 279,

27r.4,

2r1.

493

68-75.

494 75.17 Ea,l, 1'

TITS GEOMETRIES

= 4.

C* maps:

C-graph :

C-edges : 2771,

75.18

F4,1,

l ' = 1. P1 P2

-+

+P 1

2773,

2T2.

+ 3P3 + 2 P 4 + p29 p3

-P27

2775,

--f

-p33

p 4 +-p4.

C-graph : PI PI

+ P2 + 3P3 + 2 f 4 . F4,1.

C-edge : 2r,.

75.19. The Geometric Interpretation of the incidence geometries of G E A,, B,, C,,DI, as dealt with in 70.1-3, can be extended to the real case. Again the lowest dimensional linear representations of these groups are used. Compare the discussion with that in 59.7, but note that there the trunk was taken maximally compact. Gc E

B,,j,

D1.j

(i> 21,

Dl,j,*:

The representation with weights fw,and possibly 0 is used. The corresponding weight vectors x+, and possibly xo are orthonormal in the unitary inner product. C*o,= w i , DD,,x, is a multiple of x-,,Dx,a multiple yixrof xi,

75. C-GRAPHS OF INCIDENCEGEOMETRIES

495

for i = 1, ..., 1'. Since xi may be replaced by p i x i , with pi2 = yi, one may suppose that D s i = x i , i = 1, . . ., 1'. Now ( s i ,D D , , x j ) = 0 , i, j = 1,2, . . ., I ' , and the x i are real. Therefore the si.i = 1, ..., p , with 1 ~p < l', span a totally isotropic subspace of RD with respect to the invariant quadratic form; its stability group is the C-restriction of G(p,). By verifying that its stability group in the full orthogonal group of the quadratic form contains elements in each component (cf. 54.1 and note that I' < 1 is supposed) and by applying Witt's theorem (cf. 70.2) one sees that its sort consists of the totally isotropic p-dimensional subspaces. In projective terms its sort consists of the projective ( p - 1)-dimensional subspaces on the invariant quadric. Gc E A*JZ

The reprcsentation with weights wi (2w, = 0) is used. The weight vectors are correspondingly mutually orthogonal xl, . , ., x I + ]; C*wi = -w 1+2-19 i = 1, ...,1'. Hence by 59.4.1 Kxi is a multiple of x1+2-ifor i = 1, ..., 1'. Therefore the span of x I ,. . ., x, is a totally isotropic subspace for the invariant hermitean form, for p = 1, .. ., 1'. As in the preceding case, one finds that the sorts are the projective ( p - 1)-dimensional subspaces on the hermitean quadric, for p = 1,2, . . ., 1'. Gc E A,,o,*, 1 odd:

The representation with the weights wi (2w i = 0) is used; corresponding weight vectors are xI,. . ., xI+l ; C * W , ~=- ~w Z 1 ,so Dx2i-l is a multiple of x 2 i . Now G(pza)is characterized by leaving the subspace spanned by x I , . . ., x2p invariant, p = 1,2, . .., (I - 1)/2. The C-restriction of G(p2,) consists of the volume-preserving quaternion linear mappings in G that leave the p dimensional quaternion-linear subspace spanned by x I , . . ., x Z p ,with complex scalars, invariant. The sorts consist of the quaternion-projective ( p - 1)-subspaces of the quaternion-projective ( I - 1)-space, for p = 1, . . ., ( l - 1)/2. Gc

E

C L , ~Di.1:

The representation with weights iw, is used; weight vectors are x k i ; C*w2i-l = w 2 1 for i = 1, . . ., 1'. Therefore the situation is similar to that in Al,o,*.There is now, however, an invariant quaternion-hermitean or -antihermitean form; xI,. . ., x2p span a p-dimensional totally isotropic subspace The pth sort consists of the totally isotropic p-dimensional quaternion-linear subspaces with respect to the invariant form. 75.20. Historical Note The splendid theory dealt with in Sections 70-75 is due to J. Tits. It was gradually developed between 1955-1962. Almost all results were published without proof. The present reconstruction of the proof was made easier by occasional private communications by Tits. The

496

68-75.

TITS GEOMETRIES

methods are probably not very different from those intended by him. In the present exposition neither the largest generality nor the greatest economy in suppositions was aimed at. Tits' geometries are particularly important for the understanding of the exceptional groups, which have been extensively studied in the last 10 to 15 years. This theory, which has not been tackled in this book, will be the subject of a monograph. Tits' most important results on the subject of Sections 70-75 are found in Mkm. Acad. Bruxelles in 8", 29 (1955); Bull. SOC.Math. Belg. 8, 48-81 (19551957); Colloque AIgPbre Superieure C.B.R.M. (1956) Bruxelles 261-289, and particularly in Algebraic and Topological Foundations of Geometry, Colloquium 1959,175-192.

76-77 BETTI NUMBERS OF SEMISIMPLE LIE GROUPS AND REGULAR SUBALGEBRAS OF SEMISIMPLE LIE ALGEBRAS

76. AD-NILPOTENTS AND SEMISIMPLE SUBALGEBRAS OF RANK 1

Part of the results will be used in Section 77. 76.1-2

F

76.1 If a

E

E

A,.

F, a # 0, then aad-nilpotent tt $(a, a) = 0, a ad-pure tt a regular tt #(aya) # 0.

Ifa,b E F, a # 0, b # 0, then

a, b Int F-equivalent tt $(a, a) = $(b, b). This is most easily proved by assuming that F is the infinitesimal algebra of the special linear group of 2-space and applying 19.19 and 69.1. 76.2 The triple rh,eyfl of elements # O of F is called a hef-triple iff

[h,e]= e,

[h,fl= -f,

[e,fl = h.

h is the first member of some hef-triple of F iff $(h,h) = 2. e is the second (or third) member of some hef-triple iff it is ad-nilpotent #O. Given such an h, the set of its positive multiples is taken as a positive trunk of F. Then the (only) positive rootform is called A. Thus A(h) = 1. The weights of any linear representation of F with respect to this trunk are integral multiples of +A. If the representation is irreducible with topweight +PA, then its weights are the (+p - q)A, with 0 G q G p (q integral); all are simple; 0 is a weight iff p is even. 76.3-9 G E Alg Lie Com SS, A,(G) is the set of semisimple subalgebras of G of rank 1 ; 1 = rank G. F E A,(G); 9 is the representation of F in G by adjoint action, which is conducible; 9, are the irreducible components of a direct splitting of 8, acting on linear sub-spaces R, (i = I , .. ., m);dim R, = p i + 1. 491

498

76-77. BETTI NUMBERS AND REGULAR SUBALGEBRAS

76.3 A hef-triple rh,e,f 1 and an ordered trunk of F are chosen as in 76.2. Then e is also a G-ad-nilpotent. In particular, E takes a vector of weight (+pi - q)X of R , into one of weight ($pi - q + l)X, or into 0 iff q = 0. Therefore the top-weight vectors of the 8,span the kernel of E. The O-weight vectors span the kernel of k. Proposition

dim kernel E

= m(F) = number

of the R i .

dim kernel k = numberofthe Ri with evenp,.

m(F) 2 1, with the equality sign iff h is G-regular and all p, are even. The last assertion follows from the fact that dim kernel equality sign iff h is regular.

k 2 1, with the

76.4.1. Definition Fo E A,(G) is called regular if the number m(F) of Ri reaches its minimum for F = FW An ad-nilpotent e of G is called a regular ad-nilpotent, if in the set of ad-nilpotents a E G dim kernel a' reaches its minimum for u = e.

Clearly, since every ad-nilpotent is in some F following applies :

E

A,(G) (see 68.8.3), the

Proposition An ad-nilpotent e # 0 of G is contained in some regular F E A,(G) iff it is regular.

76.4.2 Let H be an ordered trunk of C and u E H such that

for all p

p(u) = 1

E

W++(G).

Clearly, u is regular and u E &. Putting e=

CI

f=

e,,

one can manage that [Gfl=

T

e-,,

pew++

EW++

z Tph, = u,

since the nodes h, span Hs,. Then Fo spanned by u,e,f belongs to A,(G) and ru,e,fl is a hef-triple of Fo. Furthermore, for any a E W * ( C ) a(u) = a(.)

=

altitude of a

(see 54.3.2). Therefore all weights of 8(Fo)are integral multiples of X and all p i are even. Hence m(Fo)= 1, and Fo is regular. This shows:

76.

AD-NILPOTENTS AND SUBALGEBRAS OF RANK

1

499

Theorem For any F the number of Riis >I; for the regular ones it equals 1. For any ad-nilpotent e, dim kernel B 2 I , with the equality sign for the regular

ones. 76.5 Let rh,e,fl be a hef-triple of F. Then h is ad-pure and therefore contained in a trunk H of G (see 17.6). An ordered dressing is assumed on H such that a(h) > 0 + a > 0 for every a E W*(G). The branch e, (with respect to H ) is a weight vector o f 6 belonging to the weight aIHnF. Proposition p(h) = 0, 3, 1 for p

E

W++(C).

Proof The p(h) are integral multiples of 3. Since p > 0, they are nonnegative. Suppose that p(h) > 1 for some p E W++(G). Then ep is for 6 a vector ofweight ph with some p > 1. Hence V;e,] is a vector of weight ( p - 1)h > 0. f can be

written as

c r,e-,

with a(h) = 1, hence with a > 0. 0 # Bepl =

c 7,[e-,,ep1

aswf

shows that for some a E W +there is a rootform p - a with ( p - a)(h)> 0, hence with p - a > 0, which contradicts the primitivity of p. This disproves p(h) > 1 and proves the assertion. An immediate consequence of the foregoing (see 76.3) is the following proposition: 76.6. Proposition For regular F E A,(G) and h E F,witht,h,(L,h) = 2, with respect to any suitably ordered trunk H of G containing h:p(L)= 1 for all p E W++(C). All h E F of all regular F E A,(G) with t,hF(h,h)= 2 are Int G-equivalent. 76.7. Proposition If rh,e, fl,rh,e, f’l are hef-triples of F,respectively, F‘ E A,(G), thenf=f’. Proof If # 0 , f - f ’ would be a weight vector of 8 of weight -A, belonging to

the kernel of 2, which, according to 76.3, should be spanned by vectors of positive weight. This showsf-f’ = 0. 76.8. Theorem F,F’ E A(G) are Int G-equivalent iff the ad-nilpotents #O of the one and the other are Int G-equivalent, iff h E F, h’ E F’, with t,hF(h,h)= t,hF,(h’,h’)= 2, are Int C-equivalent. In both cases the “only if” statement is trivial. The first and second “if” statements are proved in 76.8.1 and 76.8.2.

76-77.

500

BETTI NUMBERS AND REGULAR SUBALGEBRAS

76.8.1 F,F’may be supposed to have an ad-nilpotent e # 0 in common. hef-triples r h , e , f l , rh’,e,f” of F and F’may be assumed. Clearly, n = [e,f’ -fl E BG.

n = h’ - h E kernel 2,

With respect to the splitting G = n = C n,,

Ifp,

= 0, then

x R, under 9(associated with F),

with n, top-weight vector in R,, or = 0.

n, = 0, since n E BC. Put

x p;’ n,.

n‘ = -2

Then (exp Z ) h = h

+ n = h’,

(exp i ’ ) e = e. Thus by 76.7 (exp i ‘ ) f =f’, which proves the assertion. 76.8.2 One may suppose h = h’ and rh,e,fl, rh,e’,f’l hef-triples of F and

F‘.Let Go,GI be the 0-, respectively, l-eigenspace of k. Then EGO = B’ Go = GI. In general, iGoc GI

for all a E GI.

The a E GI, with rank i ( G O,

(7 E

sp,

W +A Ty E W-).

sp,-l

* '

spl

y < O*

Then spk-l

'

*

' spl

y

=Pk

'pk-1

Pk.

and y = sp,

' '

Conversely, any element of this form clearly belongs to I'. Therefore consists of the (k = 1, . . ., I ) 3/k = s p 1 * * ' sPk-, Pk Now with integral P k i yk

=

Pk +

Pkl

Pi,

which shows that the yk are linearly independent. Thus,

r

506

76-77.

BETTl NUMBERS AND REGULAR SUBALGEBRAS

(1 - T-')

77.2.3

Tk =

yk.

Now, from the linear independence of the yk it follows that 1 - T-I is nondegenerate ;hence 77.2.4 1 is not an eigenvalue of T.

The sum of an orbit is T-invariant; therefore by 77.2.4 it is 0, and every orbit contains positive as well as negative rootforms. Consequently : 77.2.5 every orbit intersects r a n d there are at most 1 orbits.

If there were less, some yt and y j (i # j ) would belong to the same. Then by 77.2.3 rland r jwould be equivalent under some power of T. This, however, is not possible, since both are dominant (see 33.8.1). Hence: 77.2.6 There are exactly 1 orbits. 77.3. Definition w is the eigenvalue of T with minimal positive argument and 5, is an eigenvector belonging to w . Theorem w is a primitive nth root of unity ( n = order of T). 5, is regular, that is, (&,,,a) # 0 for all a E W*.All orbits have the same length, n = (r - 1)/1. All primitive nth roots of unity occur as eigenvalues of T with a regular eigenvector. This theorem rests on the following:

5, on the basis W++ vanishes. This proposition will be proved in 77.6.

Proposition No coordinate of

Proof of the Theorem Suppose that (&,,,a)= 0 for some a E W * . Then

(t,,Ta)

=

(T-I &,,,a)= w-'(&,,, a) = 0.

5, is orthogonal to the orbit of a, hence to some Y k . Thus by 77.2.3, -T-')Tk,tw)=(Tk,(l -T)'!u)=(Tk,(l -w>tu);

7 7 . KILLING-COXETER TOOLS, BETTI NUMBERS

507

hence (rkr f,) = 0. This shows that the p,-coordinate of 4, vanishes, which contradicts the preceding proposition. Therefore 4, is regular. For some m that divides n, w is some mth primitive root of unity. 5 , is Tminvariant, as are Ref,IHBtandImf,IHst.For some real T,Re&,,;,IHat+~Im&,,:,IHst is still regular, thus situated in the interior of a chamber. It is T"-invariant, but any element of Int W* leaving an inner point of a chamber invariant is the identity (see 33.8.1). Therefore T" = 1 and, since T" = 1 and m divides n, m = n. Since on a natural basis the coefficients of any S,, hence of T, are integral, the eigenvalues of Toccur in full systems of algebraic conjugates. So any primitive nth root of unity A, as a conjugate of w , is an eigenvalue. Under the same conjugation f , passes into an eigenvector f h belonging to A, which is again regular. Suppose that the orbit of a E W * has length k . Then T k a = a, k divides n, and ( a ,5 ,)

=

(Tka , 5,)

T-k 5,)

= (a,

=~

- ~ (4,). a ,

Since 5, has been proved to be regular, it follows that wk = 1. So k is a multiple of n and, since k divides n, k = n. Now n = ( r - l)/l by 7 7 . 2 . 6 . 77.4. Theorem n Table D). Proof

(I)

=

q + I , where q is the altitude of the top rootform (see

Define the operator Z on H* by

Z commutes with every element of Int W * , which acts irreducibly on H * . Therefore Z is a scalar multiplication. From =

+m1-

t r Z = + C tr(1 - S C ( ) = + C , 2 = r - I , it follows that

( 2 ) Let p be the top rootform. Then for a

E

W+,

508

76-77.

BETTI NUMBERS AND REGULAR SUBALGEBRAS

Now every p-ladder except r-p,O,p'

( 3 ) Choose A

E

has length 1 ;thus

H,, such that p(A) = 1 for all p

E

W++. Then

4 = 4-4= P(4.

A=

(a,a)-'Aa aeW+

easily follows by the argument used in 41.6. Now

which proves the assertion. 77.5 Since Tq+' = 1, all eigenvalues of Tare powers of w . Definition k is the weakly increasing sequence of I positive integers kl G q such that the wk'provide the eigenvalues of the Killing-Coxeter tool Twith due multiplicities. The multiplicity with which a positive integer x ~ occurs q in k is called dx).

As noted in 77.3, the eigenvalues of T occur in full systems of algebraic conjugates. Hence : Theorem k l = 1 and, i f j occurs in k, then with the same frequency so does

every positive integer Q q that has the same greatest common divisor with 1 a s j ; if j occurs in k,then so does q 1 -j. In some cases this settles the structure of k.

q

+

+

E,: q + 1 = 30. There are just eight relative prime numbers mod 30; namely, 1,7,11,13,17,19,23,29.

This is the sequence k.

77. KILLING-COXETER TOOLS, BETTI NUMBERS

E, : q

+1

=

509

18. The relative primes mod 18 are

1, 5, 7 , , 11, 13, 17.

The element lacking can only be 9. E,:

q + 1 = 12. The relative primes mod 12 are

1, , 5,

3 7 ,

9

11.

Here the two lacking elements might be 3,9, or 4,8, or 6,6. F,:

q + 1 = 12. 1, 5,7, 11.

G2: q + 1 = 6 .

1, 5.

To this list one easily adds: A,:

qS1=I+l. 1 , 2)..., 1.

77.6. Proof of Proposition 77.3 A matrix presentation of T is derived by expressing the pi and T - l p , in the ' y k : Yj = SPI

*

. SP,., P i , *

Further.

By 77.6.1-2 Tcan be written on the ordered basis p l , . . ., pI as the product of an upper and a lower triangular matrix. Replacing p j , y j by p;

= (pji

pj)-''* pj,

one gets a more symmetric expression : I-

I

7; = (pj, pj)-1'2 y j ,

510

76-77.

BETTI NUMBERS AND REGULAR SUBALGEBRAS

where the ajk = ( p j , P J ) - ’ ’ ~( f k , p k ) - ” 2

(pj, f k )

are the cosines of the angles of the rootforms. They form a maxtrix A ‘Ujk’:,k, 1 with ajj

= 1,

ajk

90

a,k # 0

=

for j # k , iff p,,

pk

are connected in the graph.

Let B be the lower triangular matrix arising from A by putting

bjk= k

for j > k ,

j k

=1

for j = k,

=O

for j c k.

Then, on the ordered basis p;,

..., pi, the matrix of T-’ is -Bf-l B

(where B‘ is B transposed), that of Tis -B-’B‘. Note that A

=t ( B

+ B‘).

The eigenvalues X of Tare determined by the equation det $(B’ + AB) = 0. The matrix CA = $(B’

+ hB)

is real, symmetric, and positive definite for h = I (since the inner product is so on HZ). For any h with Ihl = 1 the reality and symmetry will now be restored by multiplying the ith row byfi g, the ith column byf;’, such that one gets fhbjk(fjg)fL1

= a.lk

for

3bkj(fJg)f;’

=ajk

for k > j ,

+ wh)f;l

real.

To reach this goal one must fulfill the conditions

<j,

77. KILLING-COXETER TOOLS, BETTI NUMBERS 77.6.3

fjf? g = 1

for j < k

=A-'

for j >k

51 1

for every pair j , k such that pj,pk are connected in the graph, and 77.6.4

+(l

+ A) g

real.

Instead of 77.6.3 one can write

& f;'

77.6.5 f i fi' g = A-',

g

=

for j > k and p,, pk connected.

1

Thus one must postulate g2 = A-'. Indicating the square root of X with a nonand its inverse by A-I12, one puts negative real part by = A-"2

Thus +(1

+ A) g = +(A112 + A-112)

fulfills 77.6.4. Now 77.6.5 takes the form 77.6.6

S, = X-1/2&

for j > k and p j , Pk connected.

This can be fulfilled with nonzero numbers, since the graph is a tree. By this modification a matrix D, has arisen from CA.It agrees with A outside the main diagonal, whereas in the main diagonal it bears the value The condition det(T - A) = 0 is equivalent to det C, = 0, which is equivalent to det D A= 0, that is,

+

det(A - (1

-

+(All2

+

= 0.

The eigenvalues a of A correspond to eigenvalues A of T by means of cr = 1 - +(A112

+ A-'/2),

and the corresponding eigenvectors differ in every coordinate by a factor # 0. For this one need only look at the construction of DA and observe that the kernels ofthe linear transformations with matrices X + B-I B' and C, = B' + XB all coincide. The minimal eigenvalue of A , called ao, corresponds to the eigenvalue of T with a minimal positive argument. Since A is positive definite, A - ct0 is still positive semidefinite, which implies that the corresponding eigenvector has no vanishing coordinate, thanks to the following : Lemma Let Q be a real positive semidefinite quadratic form, b , , ..., b, a basis that does not admit any proper partition in mutually orthogonal subsets, and ufj G 0 for i # j i f rufjlf,j,I is the matrix of Q on the ordered basis b , , . . ., b,. Then no coefficient on this basis of an arbitrary nonzero isotropic element vanishes.

512

7677. BETTI NUMBERS AND REGULAR SUBALGEBRAS

Proof After a rearrangement of the indices it is sufficient to refute the

existence of numbers x l , ..., xi and a positive integer m less than I with

c U,,XJ

=

J

-

XI

0,

* *

x, # 0,

X,+l

=

* * *

= 0.

Such a system, if it existed, would imply

0=

x

UtJXiX,

2

I J

c 4,

lXil

*

Ix,I 2 0,

where the first 2 sign is justified by ul, G 0 for i # j and the second because of the positive semidefiniteness. Thus,

c

U i J IXiI * IXJI =

i.J

0;

hence, again because of the semidefiniteness,

c4 1 J

IXJl = 0

and

For i > m all summands are G 0, thus = 0. Hence u,, = 0 for i > m, j G m . But this means that the quadratic form breaks in a way that is excluded by assumption. It is easily seen from the connectedness of the graph that the lemma applies to the proof of the proposition. 77.7-1 4. Killing-Coxeter Elements 77.7 According to 33.1 T transferred to H i s the restriction on H of some I with t E G. If G is given in third dressing on H, t may even be supposed to belong to the corresponding Gun,Such a t is called a Killing-Coxeter element. More precisely : Definition For any ordered third dressing of G on any ordered trunk H with, moreover, any ordering of the primitive rootforms, t is called a KillingCoxeter element if t E Gunand I I , = T. Remember that I is pure. Definition For any Gun,a is called regular if a a" - 1 is minimal (hence l ) .

E

C,, and the nullity of

Theorem Let t be a Killing-Coxeter element inducing T and let L, be the span of the earwith u. in the orbit of yi,i = 1, . . ., 1 (see 77.2.1). Then

77.

KILLING-COXETER TOOLS, BETTI NUMBERS

77.7.1 the eigenvalues of fl,, are the q multiplicity 1,

513

+ 1 different powers of w , each with

77.7.2 the eigenvalues of f are the components of the sequence k, and the q + 1 different powers of w , each with multiplicity I , 77.7.3 f has order q

+ 1,

77.7.4 t is regular, 77.7.5 the Killing-Coxeter elements corresponding to the same T are conjugate.

In 77.10, all Killing-Coxeter elements will be shown to be conjugate. Proof 77.7.4 is an immediate consequence of 77.7.1 and the fact that fhas no eigenvalue 1 on H.

77.7.2 follows from 77.7.1. Let F be the 1-eigenspace of f acting on G and FI = F n L,. Since fe, is a scalar multiple of era, the branches can be supposed renormed (in first dressing) such that fp ey,= eTPyl for 0 < p < q. Then Fl, if # {0}, is spanned by 4

z

eTpy~,

p=o

hence dim FI s 1. Clearly

F E Z Fi, and dim F 2 1. Since by 77.2 the number of orbits is 1, dim Fl = 1, which implies dim F = I , and fa+' eyl = e,,,,

so the order of f is q + 1, which proves 77.7.3.

514

76-77.

BETTI NUMBERS AND REGULAR SUBALGEBRAS

The eigenvectors of Iin Liare the e w-~P

T pey,

p=o

belonging to the eigenvalues wk (k = 0, 1, ...,q), respectively. This proves 77.7.1. If t’ is another Killing-Coxeter element, then f’ = exp k . I with some h E H. One must look for an hl E H such that exp h, I . exp(-h,) = exp h *I This , condition is fulfilled by an hl with h, - ThI = h, which, according to 77.2, exists. This proves 77.7.5. 77.8 In the sequel ordered second dressing on another trunk H’ will be needed. One adopts the following convention : Notation Rootforms, second dressing branches, and so on, with respect to H’, are distinguished from those with respect to H by a prime.

77.9 Elements of G of finite order belong to some maximal compact subgroup, hence to some Gun with a suitable Cun.U p to conjugacy they may be supposed to be in Gun. Theorem All regular elements of G of order q regular element of G has order G q.

+ 1 are conjugate.

No

Proof The given regular element is of the form exp u with u in the principal domain (see 31.4) of some suitable suitably ordered trunk HLn. Thus, if P‘

=

c 41 P;

is the top rootform, Im ~ ’ ( u-=) 27r.

Im p;(u) > 0,

If exp u has order q + 1, all a’(u) (a’E W*’)are integral multiples of 2ni/(q + 1);

lm p;(u) =pi * 27r/(q + 1) Jm P ’ W

=

withpositiveintegersp,.

c 4i Im p ; w x Pi41 W =

’

q + 1).

This must be O,

cPi% q, C, is defined as (0); furthermore Go= H ' . u' acts on C, as the multiplication by j . G = Z C,.

Clearly dim Cj is the number of rootforms of altitudej. As the respective multiplicities of the eigenvalues j of u', the dim Cj are closely connected to the multiplicities of the eigenvalues of T = exp(2~ri/(q+ 1))u'. The latter are known from Theorem 77.7. This leads to the

+ 1 and with due account to multiplicities the eigenvalues of u' are the components of the sequence k and 1 times each of the numbers 0, 1, . .., q. In other words,

Theorem After reduction mod q

d i m G j + d i m G j ~ , ~ , = d i m G , + d i m G , , , ~ , = 1 + ~ ( j ) ( j = O ,..., q ) , if ~ ( is j )the multiplicity o f j in the sequence k . Since the dim C jcan be read from Table D, this establishes the sequence k . 77.12 The Sequence k

A, : B, :

1,2, . . ., I 1,3, . . ,,21 - 1 (the odd numbers)

76-77.

516

c1: D, (I even) : D, (I odd): E, : E,: Es : F, : G2 :

BETTI NUMBERS AND REGULAR SUBALGEBRAS

1,3, . . ., 21 - 1 (the odd numbers) 1,3, . . ., I - 1, I - 1, . .,, 21 - 5,21- 3 (the odd numbers, with the multiplicity 2 for I - 1) 1,3, ,.,,I - 2 , l - 1, I, . .., 21 - $21 - 3 (the odd numbers and the number I - 1 ) 1,4,5,7,8,11 1,5,7,9,11,13,17 1,7,11,13,17,19,23,29 1,5,7,11 195

An inspection of this table shows: 77.13

dim G,

+ dim G,+,-,

=

for j

1

=

0, . . ., q + 2.

Compared with 77.1 1 , this yields the following: 77.14. Theorem qJ(j)=dimG,-dimG,+, for j = O , ...., q ; in other words, ~ ( jis) the drop in the number of rootforms of a given altitude x when passing from x = j to x = j 1. For this formula, which makes the computation of k much easier, one can provide an interesting general proof:

+

Proof H,

=H

belonging to

UJ'

n (G,

+ G,-,,-,) is the eigenspace in H of exp(2ri/(q + 1))u'

= uJ-4-I.

Thus, dim H,

(1)

=

and

Ho = (01 by 77.2. Corresponding to the former 4, in H* there is a regular x E H I . Put

and define Gaol G,o, and GSoanalogously. Call u+ and u- the projections onto G,, and G, 59.1 [...I, 0.4, 2.1 Il...ll, 36.13 [...,...1, 72, 7.3 [...,...I, 72.2

AUTHOR INDEX [Occurring in notions.]

(Occurring in theorems and formulas.)

Ado, 1. D., (7.3) Banach, S., 34.4 Berger, M., 64.12 Betti, E., [77.1] Borel, A., [68.2],77.1.6,77.1.10 Bott, R., 77.1.10 Bourbaki, N., 70.2 Brauer, R., 49.9, 77.1.4 Brown, G., @ I S ) , 21.6 Bruhat, F., (68.9) Cartan, E., 20.14, 26.25, 32.1 1, 33.17, 35.10, 38.6,45.8,49.9,51.23,57.9,58,59.9,60.24, 62.10, 64.12, 65.6, 67.4, 77.1.2 Cartan, H., 77.1.6 Casimir, H. B. G., [40.1], 40.4, 50.12 Chevalley, C., [27.3.2], 27.3.4, 77.1.2, 77.1.5-6, 77.1.8- 10, 77. I5 Clifford, W. K., [49.2] Coleman, A. J., 77.28 Coxeter, H. S. M., 26.25, 77.1.9-10, [77.2], [77.7], 77.28 Dieudonne, J., 59.7 Dynkin, E. B., 26.25 Ehresniann, C., 77.1.4 Eilenberg, S., 77.1.2 Est, W. T. van, 12.17 Freudenthal, H., 4.8.5, 10.7, 11.8, 12.17, 26.25, 30.6, 33.17, 37.6, 45.8, 48.8, 49.9, 50.12, 51.23, 57.9, 60.24,68.15 Frobenius, G., 37.6 Gantniakher, F., 18.7, 51.23 Gotd, M., 12.17 Haar, A., [34.2], 34.4 Harish-Chandra, 26.25, 40.4, 45.8 Helgason, S., 64. I 2 Helmholtz, H. von, 63.10 Hilbert, D., 6.5, 77.1.9 Hopf, H., 63.10, 77.1.4-5 Hurwitz, A., 35.10 Iwasawa, K., 61.5 Jacobi, C. G. J., [7.2], 7.8 Jacobson, N., 68.15 Karpelcvif, F. I . , 68.15 Killing, W., L14.41, 20.14, 26.25, [77.2], [77.7], 77.28 Klein, F., 63.10

Kneser, H., 10.7 Kostant, B., 76.10, 77.1.10, 77.28 KOSZU~, J.-L., 77.1.5 Leray, J., 77.1. I , 77. I .5 Levi, E. E., (50.2) Lie, S., 16.11, 6.5, [7.31, (13.9), 20.14. 63.10 Lipschitz, R., 49.9 Molien, Th., 77.1.10 Montgomery, D., 6.5 Morozov, V. V., 68.15 Myer, S. B., 63.10 Neuniann, J. von, I I .8, 34.4 O'Meara, 0. T., 70.2 Peter, F., 35.10, 37.6 Poincare, H., 77.7.1, (77.11 Pontrjagin, L. S., 77.1.4 Rham, G. de, 77.1.1 Rinow, W., 63.10 Russell, B., 0.1 Saks, S., 34.4 Samelson, H., 77.1.5 Schafer, R. D., 26.25 Schlafli, L., 26.25 Schur, F., 10.7 Schur, I., 35.10, (36.2), 37.6 Seminaire Sophus Lie, 39.8 Shapiro, A., 77.1.10 Shephard, G. C., 77.1.9, 77.28 Steenrod, N., 63.10 Steinberg, R., 77.1 .lo, 77.28 Tits, J., 26.25, 68.15, [71.2], 75.20 Todd, J. A,. 77.1.9, 77.28 TBgd, S., 19.23 Verma, D.-N., 57.7, 77.28 Vries, H. de, 17.19, 19.23, 26.25, 51.23, 57.7, 60.24, 68. I5 Wacrden, B. L. van der, 26.25, 50.12 Weil, A., 34.4, 77.1.6 Weyl, H., r21.41, 21.6, [23.1], 24.5. 31.13, 32.11, 33.17, 35.10,37.6, r41.71, (47.7-8), 47.9, 49.9, 5 1.23 Whitehead, J. H. C., 50.12 Witt, E., 39.8, (70.2) Yen Chi-Tah, 77.1.5 Zippin, L., 6.5

547

Pure and Applied Mathematics A Series of Monographs and Textbooks Edited by

Paul A. Smith and Samuel Ellenberg Columbia University, New York

1 : ARNOLD SOMMERFELD. Partial Differential Equations in Physics. 1949 (Lectures on Theoretical Physics, Volume VI) 2: REINHOLD BAER.Linear Algebra and Projective Geometry. 1952 3: HERBERT BUSEMANN A N D PAUL KELLY.Projective Geometry and Projective Metrics. 1953 4: STEFAN BERCMAN A N D M. SCHIFFER. Kernel Functions and Elliptic Differential Equations in Mathematical Physics. 1953 BOAS,JR. Entire Functions. 1954 5: RALPHPHILIP 6: HERBERT BUSEMANN. The Geometry of Geodesics. 1955 7 : CLAUDECHEVALLEY. Fundamental Concepts of Algebra. 1956 8: SZE-TSENHu. Homotopy Theory. 1959 9: A. M. OSTROWSKI. Solution of Equations and Systems of Equations. Second Edition. 1966 10 : J. DIEUDONN~. Treatise on Analysis. Volume I, Foundations of Modern Analysis, enlarged and corrected printing, 1969. Volume I1 -in preparation 11: S. I. GOLDBERG. Curvature and Homology. 1962. 12 : SICURDUR HELCASON. Differential Geometry and Symmetric Spaces. 1962 Introduction to the Theory of Integration. 1963. 13 : T. H.HILDEBRANDT. 14 : SHREERAM ABHYANKAR. Local Analytic Geometry. 1964 15: RICHARD L. BISHOPA N D RICHARD J. CRITTENDEN. Geometry of Manifolds. 1964 16: STEVEN A. GAAL.Point Set Topology. 1964 17: BARRYMITCHELL.Theory of Categories. 1965 18: ANTHONYP.MORSE.A Theory of Sets. 1965

Pure and Applied Mathematics A Series of Monographs and Textbooks

19: GUSTAVECHOQUET. Topology. 1966 20: 2. I. BOREVICH A N D I. R. SHAFAREVICH. Number Theory. 1966 A N D JUAN JORCE SCHAFFER. Linear Differential Equations 21 : Josh LUIS MASSERA and Function Spaces. 1966 22 : RICHARD D. SCHAFER. An Introduction to Nonassociative Algebras. 1966 23: MARTINEICHLER.Introduction to the Theory of Algebraic Numbers and Functions. 1966 24 : SHREERAM ABHYANKAR. Resolution of Singularities of Embedded Algebraic Surfaces. 1966 25 : FRANCOIS TREVES. Topological Vector Spaces, Distributions, and Kernels. 1967 26: 27: 28 : 29 : 30: 31 : 32 : 33: 34: 35 : 36 :

PETER D. LAXand RALPHS. PHILLIPS. Scattering Theory. 1967 OYSTEINORE.The Four Color Problem. 1967 MAURICE HEINS.Complex Function Theory. 1968 R. M. BLUMENTHAL A N D R. K. GETOOR. Markov Processes and Potential Theory. 1968 L. J. MORDELL. Diophantine Equations. 1969 J. BARKLEY ROSSER.Simplified Independence Proofs : Boolean Valued Models of Set Theory. 1969 WILLIAMF. DONOGHUE, JR. Distributions and Fourier Transforms. 1969 MARSTONMORSEA N D STEWART S. CAIRNS.Critical Point Theory in Global Analysis and Diff ereritial Topology. 1969 EDWINWEISS. Cohomology of Groups. 1969 HANSFREUDENTHAL A N D H. DE VRIES.Linear Lie Groups. 1969 LASZLO FUCHS. Infinite Abelian Groups : Volume I. 1970 In firefiaration KEIO NACAMI.Dimension Theory. PETER L. DUREN.Theory of HP Spaces.

This Page Intentionally Left Blank